JP2002032766A - Image recognition apparatus and method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To recognize an object having a specific shape from an input image using a model image, but also recognizes an object which has a shape to be detected but is not in the model image. SOLUTION: An image database 1 for registering a shape identifier representing a shape of a detection target and an image of an object having the shape as a model image in advance, a model generating unit 2 for extracting a shape feature from the model image for each shape, Shape information database 3 in which shape features and shape identifiers are stored in pairs.
An image input unit 4 for inputting an image for detecting an object, an image cutout unit 5 for cutting out the input image as a partial image,
A shape classification unit 6 for comparing a partial image with a shape feature to determine whether there is a shape to be detected, and an output unit 7 for outputting information representing the shape and position information of the shape in an input image. Thus, a specific shape and its position can be detected and output with high accuracy from an input image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention extracts shape features to be detected in advance from a large number of model images obtained by photographing an object having a specific shape, and compares features extracted from an input image with features of the model. Accordingly, the present invention relates to an image recognition apparatus and method for detecting an object having a specific shape from an input image and outputting the position and shape thereof.

[0002]

2. Description of the Related Art A conventional image recognition apparatus is disclosed in
One described in Japanese Patent No. 5140 is known.

FIG. 17 shows a block configuration of a conventional image recognition device, and a display device 101 for displaying an image,
A main controller 102 for controlling the overall operation, an internal memory 103 for storing operation programs and the like, a disk 104 for storing reference patterns and the like, a television camera 105 for photographing products such as samples and identification objects, and a television camera An image input unit 106 for converting the image of the product photographed at 105 into digital image information, and an image rotating unit 107 for performing a positioning process for turning the product image in a certain direction for each category of the gray image converted to digital image information. And an image information extraction unit 10 that samples the rotated image at a certain rate and extracts the grayscale value of the sub-sampled image as a feature.
8, a dictionary creation unit 109 that includes an average vector calculation unit 109a that calculates an average vector for each category from the obtained features and creates a dictionary (reference pattern) including the average vectors, and an identification target whose category is unknown. And a vector distance comparison unit 110a for extracting an average vector having the shortest distance from the vector from the average vector of the dictionary creation unit 109. The identification unit 110 identifies an unknown identification target. The image input unit 106,
Image rotation unit 107, pixel information extraction unit 108, identification unit 11
It comprises a parameter setting unit 111 for optimizing the parameter of 0.

[0004]

In such a conventional image recognition apparatus, when attempting to recognize an object by its shape and classify the shape, all the images of the object having the same shape and greatly different grayscale values are classified into one category. , The recognition becomes difficult, and it is necessary to create one category using images having similar grayscale values. Therefore, there is a problem that the number of categories is increased and processing time is required.

SUMMARY OF THE INVENTION The present invention solves the above-mentioned conventional problems.
The purpose is to quickly recognize the shape of an object in an image and output its position.

[0006]

In order to solve the problem, a shape identifier representing a shape of a detection target and an image database in which an image of an object having the shape is registered in advance as a model image, A model generating means for extracting shape features in advance from the model image, a shape information database for storing the shape features and the shape identifiers in advance, and an image input unit for inputting an image for detecting an object. An image cutout unit that cuts out a part of an input image as a partial image, and a shape classification unit that determines whether a detection target shape is present in the partial image by comparing the partial image with the shape feature, If the input image has a shape that matches the shape to be detected, an output unit is provided that outputs information indicating the shape and information indicating the position of the shape in the input image. Those were.

According to the present invention, by extracting a common shape feature from a large number of model images in advance and comparing the shape feature with the input image, the presence / absence of an object in the input image can be determined at a high speed with a small amount of data. , And the position and the detected shape can be output.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS According to a first aspect of the present invention, in a device for detecting an object having a specific shape from an image, a shape identifier representing a shape of a detection target and an image of the object having the shape are modeled. An image database pre-registered as an image, model generation means for pre-extracting shape features from the model image for each shape, and shape information pre-stored in pairs of the shape features and the shape identifiers A database, an image input unit for inputting an image for detecting an object, an image cutout unit for cutting out a part of the input image as a partial image, collating the partial image with the shape feature, and detecting a detection target in the partial image. A shape classifying unit that determines whether or not there is a shape, and an output unit that outputs information indicating a shape of the detected detection target and information indicating a position of the shape in an input image. Those in extracts shape feature representing a number of model image, by comparing the input image with the shape feature has the effect of detecting a shape from an input image at high speed with a small amount of data and outputs the position.

According to a second aspect of the present invention, in the image recognition apparatus according to the first aspect, the model generating means includes, for each shape, an average image of the model images belonging to the shape and a model of each pixel of the model image. Extract the variance as shape features,
The average image, the variance, and the shape identifier are paired and stored in the shape information database for each shape. This has an effect that the shape of the input image can be compared with the shape of the model with high accuracy.

According to a third aspect of the present invention, in an apparatus for detecting an object having a specific shape from an image, a shape identifier representing a shape of a detection target and an image of the object having the shape are registered in advance as a model image. An image database and a basis vector of the feature space are calculated from the pixel values of all the model images, all the model images are projected as a model image vector in the feature space, and each model image vector to which the shape identifier is assigned is Calculating a characteristic statistic for each shape from the model image vector as a shape feature parameter and assigning the shape identifier; and a base vector and a shape feature parameter from the model generation unit together with the shape identifier. A shape information database to be registered, an image input unit for inputting an image for detecting an object, and a part of the input image An image cutout unit that cuts out as a partial image, the partial image is projected onto the feature space using the basis vector to obtain a partial image vector, and the partial image vector is compared with a model using the shape feature parameter to match the model. A shape classification means for determining whether or not there is a shape to be detected in the partial image, and information indicating the shape of the input image if there is a shape that matches the shape of the target and information in the input image of the shape. And an output unit that outputs information representing the position of the image. By comparing the input image with the model shape in the feature space, efficient detection can be performed and the processing speed is high.

According to a fourth aspect of the present invention, in the image recognition apparatus according to the third aspect, the calculation of the shape characteristic parameter for each shape by the model generating means is performed by using the model image calculated from a model image of an object having the same shape. It calculates shape characteristic parameters for each shape from the average vector and the covariance of the vector, and has an effect that a specific shape can be efficiently and accurately detected from an input image.

The invention described in claim 5 is the invention according to claim 3 or 4.
In the image recognition device described in the above, the model generation means calculates an average image of the model images for each shape, calculates a base vector from pixel values of all the average images, and stores all the model images in a feature space. Projecting as a model image vector and assigning the shape identifier to each model image vector. By generating a feature space that distinguishes images representing each shape well, it has an effect that similar shapes can be well classified. .

According to a sixth aspect of the present invention, in the image recognition apparatus according to any one of the third to fifth aspects, the shape identifier includes information on which part of the object the shape is. Even if a part of the object is concealed, an object having a specific shape can be detected only by the shape of the part.

According to a seventh aspect of the present invention, in the image recognition apparatus according to the sixth aspect, the shape classification means estimates at least one region of the entire object in the input image for each shape identifier of the determined partial shape. In order to output the position of the object region by summing the region of the entire object estimated for each partial image of the input, even if a part of the object is hidden, in both the shape of the part and the entire shape, This has the effect that an object having a specific shape can be detected with high accuracy.

According to an eighth aspect of the present invention, in the method for detecting an object having a specific shape from an image, a shape identifier representing a shape to be detected and an image of the object having the shape are registered in advance as a model image. A database, a step of pre-extracting a shape feature from the model image for each shape, a step of inputting an image for detecting an object, and a shape information database in which the shape feature and the shape identifier are stored in pairs. Cutting out a part of the input image as a partial image, collating the partial image with the shape feature, and determining whether or not there is a shape to be detected in the partial image, Outputting information representing the shape of the detected object and information representing the position of the shape in the input image. Extracting features, by comparing the input image with the shape feature has the effect of detecting a shape from an input image at high speed with a small amount of data and outputs the position.

Hereinafter, an embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIG.

(Embodiment 1) FIG. 1 is a block diagram of an image recognition apparatus according to Embodiment 1 of the present invention.

In FIG. 1, reference numeral 1 denotes a gray-scale image of a large number of objects having a shape to be recognized, and a shape name, an image file name, and upper left corner coordinates and lower right corner coordinates of a rectangular area circumscribing the object in the image. An image database 2 stored together with the identifiers, 2 is a model generating means for inputting all the grayscale images of the shape for each shape to be recognized from the learning database 1 and extracting features, 21 is all the grayscales input from the learning database 1 A feature amount extraction unit that calculates an average of each pixel in a rectangular area circumscribing the object and a variance from the average for each shape to which the image belongs, and outputs the average along with a shape identifier;
Is a shape information database in which the average image, variance, and shape identifier for each shape are input and stored from the feature amount extraction unit 21; 4 is an image input unit for inputting an image whose shape is to be determined; and 5 is a shape information database. 3 is an image cutout unit that inputs the shape identifier stored in 3 and cuts out an image having the same size as the shape of the detection target from the input image. 6 indicates whether there is a shape of the detection target in the image input from the image cutout unit 5. A shape classifying unit 61 for comparing the image input from the image cutout unit 5 with a shape feature of a model extracted from the shape information database 3 to determine whether or not the image matches a shape to be detected; Reference numeral 7 denotes a display or the like which displays the detected shape and the position in the input image when the result that the shape to be detected exists in the input image is input from the shape classifying means 6 Which is an output unit for displaying.

FIG. 2 is a block diagram showing a case where the image recognition apparatus is realized by a computer. Reference numeral 201 denotes a computer; 202, a CPU;
Reference numeral 4 denotes a keyboard and display; 205, a storage medium unit such as an FD, PD, MO, or DVD for reading an image recognition program; 206 to 208, an I / F unit; 209, a CPU bus; , 211 is an image database for capturing previously stored images, 212 is a shape information database storing model images of objects of various shapes together with shape identifiers indicating the shapes, and 213 is the shape of the obtained object. And an output terminal for outputting the position via the I / F unit.

The operation of the thus configured image recognition apparatus will be described below with reference to the flowcharts of FIGS. FIG. 3 is a flowchart showing the operation of the model generating means, FIG. 4 is a flowchart showing the operation from input of image data to be recognized to output of a recognition result, and FIG. 5 is stored in the image database 1. FIG. 6 shows an example of an average image of one shape and a shape identifier stored in the shape information database 3, and FIG. 7 shows an image cutout unit 5 for an image input from the image input unit 4. FIG. 8 shows an example of a region to be cut out as a rectangle.
Is an example in which the result detected by the shape classification means 6 is output to the display by the output unit 7.

Before performing recognition, data of shape information used at the time of recognition is constructed in advance. In the image database 1,
A gray image file of the object as shown in FIG. 5 is stored as a model image. Each image is provided with a shape of the object in the image, an image file name, and a shape identifier that describes an upper left corner coordinate and a lower right corner coordinate of a rectangle circumscribing the object as an area of the object. FIG. 5 is an example of a model image of a sedan in which the angle with respect to the camera that captured the image and the distance from the camera are constant.

When a vehicle having the shape of the sedan a as shown in FIG. 5 is to be detected, the model generating means 2 inputs from the image database 1 all the model images whose shape names are sedan a together with the shape identifier. , Feature amount extraction unit 21
Then, an average image of the rectangular area images indicated by the target area is obtained (step 301). At this time, since each model image is an image of an object having the same shape, all the rectangular areas indicated by the target area of each model image have the same size, and the average image also has this size.

In the case of the sedan a shown in FIG. 5, the average image size is 148 pixels horizontally and 88 pixels vertically. Next, the feature amount extraction unit 21 calculates the variance in each pixel from the pixel value of each pixel of the rectangular area image indicated by the target area of all the model images and the pixel value of the corresponding pixel of the average image ( Step 302).

Finally, the feature quantity extraction unit 21 uses the average image, shape identifier, and variance value of each pixel of the sedan a as shown in FIG. It is stored in the shape information database 3 (step 303). When there are a plurality of detection targets, the processes from step 301 to step 303 are similarly performed for other shapes.

When recognizing the sedan a, first, image data to be recognized is input from the image input unit 4 (camera 210 or image database 211) (step 4).
01). As shown in FIG. 7, the image cutout unit 5 successively cuts out a partial image of the same size as the average image of the sedan a stored in the shape information database 3 while moving an arbitrary pixel from the input image (step 402). .

The shape classifying means 2 inputs one partial image from the image clipping unit 5 and outputs
Input the average image of the sedan a and the variance of each pixel from
In the partial shape classification unit 61, for each pixel of the partial image, the square of the pixel value difference from the corresponding pixel of the average image is obtained, divided by the variance, and the sum over all the pixels is obtained. The distance from the average image is set (step 403). When there are a plurality of shapes to be detected, the partial shape classifying unit 61
Performs step 403 for each shape (step 40
4) If the minimum value of all the distance values is smaller than a certain value (step 405), it is determined that the partial image includes an object having the shape of the average image for which the minimum distance has been calculated (step 406). .

If the minimum value of the distance is equal to or more than a certain value, it is determined that no object exists in the partial image (step 4).
07). The partial shape classification unit 61 performs up to step 407 for all partial images cut out from the input image.
(Step 408) If there is a partial image determined to have an object, the output unit 7 displays the shape of the object at the position of the partial image of the object in the input image as shown in FIG. Are superimposed and output (step 409). This output is output from the output terminal 213 via the I / F unit 208.

(Embodiment 2) FIG. 9 shows a block diagram of an image recognition apparatus according to Embodiment 2 of the present invention.

In FIG. 9, reference numeral 1 denotes an upper left corner of a rectangular area circumscribing a shape name, an image file name, and a shape to be detected in an image, the same size regardless of the shape. An image database that stores angular coordinates and lower right corner coordinates together with shape identifiers, 2 is a model generation unit that inputs all grayscale images of the shape to be recognized from the learning database 1 and extracts features, and 20 is a learning database 1 A feature space generation unit that generates a feature space from all the model images input from, outputs its base vectors to a shape information database, and projects all model images as feature image vectors into the feature space; Find the average and variance of model image vectors for each shape from all input model image vectors and output them along with shape identifiers Symptoms amount extracting unit, 3 feature space generating unit 2
A shape information database that inputs a base vector of a feature space from 0, and inputs and stores an average and a variance of model image vectors for each shape together with a shape identifier from a feature amount extraction unit 21; An image input unit 5 for inputting an image is an image cutout unit for inputting a shape identifier stored in the shape information database 3 to cut out an image having the same size as the shape to be detected from the input image. A shape classification means 60 for determining whether or not there is a shape to be detected in the input image, and a partial image input from the image cutout unit 5
A feature space projection unit for projecting a feature image as a partial image vector into the feature space using the basis vector input from the image processing unit; and 61 each of an average of the partial image vector input from the feature space projection unit 60 and the model image vector input from the shape information database 3 The partial shape classification unit 7 for determining the distance to the shape of the object to be detected and determining whether or not it matches the shape of the object to be detected is detected when the result that the object to be detected is present in the input image from the shape classifying means 6 is detected. An output unit that displays a shape and a position in the input image on a display or the like.

The operation of the thus configured image recognition apparatus will be described below with reference to the flowcharts of FIGS. FIG. 10 is a flowchart showing the operation of the model generation means, FIG. 11 is a flowchart showing the operation from input of image data to be recognized to output of a recognition result, and FIG. 12 is stored in the image database 1. 7 shows an example of a model image, FIG. 7 shows an example of a region to be cut out by the image cutout unit 5 with respect to the image input from the image input unit 4, and FIG. 8 shows a result detected by the shape classification unit 6 in the output unit. 7 is an example of outputting to a display.

Before performing recognition, data of shape information used at the time of recognition is constructed in advance. In the image database 1,
A gray image file of an object as shown in FIG. 12 is stored as a model image. Each image is given a shape identifier describing the shape of the object in the image, the image file name, and the upper left corner coordinates and lower right corner coordinates of the rectangle circumscribing the detected shape as the object area. FIG. 12 is an example of a model image of a sedan and a model image of a bus in which the angle with respect to the camera that captured the image and the distance from the camera are constant.

When a vehicle and a bus having the shape of the sedan a as shown in FIG. 12 are to be detected, the model generating means 2 determines from the image database 1 that the model name of the shape identifier of the sedan a and the bus name are the model image of the sedan a. All model images on the back are input to the feature space generation unit 20 together with the shape identifier.

The feature space generation unit 20 obtains eigenvalues and eigenvectors from the pixel values of the rectangular area image indicated by the target area (step 1001). At this time, the rectangular areas indicated by the target areas of each model image are all the same size, and are 148 pixels horizontally and 88 pixels vertically in FIG. In the case of the bus in FIG. 12, the same rectangular area from the images of all the buses is set as the target area. When calculating the eigenvector, first, a vector is created by arranging pixel values in a line for each model image, and then a vector is created by subtracting the average vector of all the vectors from each vector, and the eigenvalue is calculated. And guide the eigenspace.

The feature space generator 20 stores eigenvectors corresponding to the n largest eigenvalues in the shape information database as base vectors (step 1002).
All the model images are projected to the eigenspace using the n eigenvectors to obtain model image vectors (step 10).
03).

The feature amount extraction unit 21 is provided with a feature space generation unit 20
, The model image vector and the shape identifier are input, and the average and covariance of the model image vectors having the same shape identifier are obtained (step 1004). The feature amount extraction unit 21 stores the average of all model images, the average of model image vectors for each shape, the covariance, and the shape identifier in the shape information database 3 (step 1005).

When performing recognition, first, image data to be recognized is input from the image input unit 4 (step 11).
01). The image cutout unit 5 includes the shape information database 3
The size of the model image is calculated from the target area of the shape identifier stored in the input image, and partial images of the same size as shown in FIG. 7 are sequentially cut out from the input image while moving any pixels (step 1102).

The shape classifying means 2 inputs one partial image from the image cutout unit 5 and outputs
, And the feature space projection unit 60 calculates
Projects a partial image into the eigenspace as a partial image vector
(Step 1103). The partial shape classifying unit 61 receives the partial image vector from the feature space projecting unit 60 and the average vector and covariance of each of the sedan a and the bus from the shape information database 3, and inputs the partial image vector and the Mahalanobis of each average vector. Calculate the distance (Step 110
4).

If the minimum value of all Mahalanobis distance values is smaller than a predetermined value (step 1105), it is determined that the partial image has an object having a shape having an average vector for which the minimum distance has been calculated (step 1105). 1106). If the minimum value of the distance is equal to or more than the predetermined value, it is determined that no object exists in the partial image (step 1107). The feature space projection unit 60 and the partial shape classification unit 61 perform steps 1103 to 1107 for all partial images cut out from the input image (step 1108). If there is a partial image determined to have an object, As shown in FIG. 8, the output unit 7 outputs a diagram representing the shape of the object at the position of the partial image of the object in the input image (step 1109).

This output is supplied to the I / F unit 208
Is output from the output terminal 213 via the.

(Embodiment 3) FIG. 13 is a block diagram of an image recognition apparatus according to Embodiment 3 of the present invention.

In FIG. 13, reference numeral 1 denotes a gray-scale image of a large number of objects having a shape to be recognized, divided into rectangular partial areas of the same size for each image, and the shape name of each part, the image file name, and the like. An image database that stores shape identifiers describing the upper left corner coordinates and lower right corner coordinates of the rectangular area of the portion, and a model that inputs all grayscale images of the shape to be recognized from the learning database 1 and extracts features The generating means 20 generates a feature space from the pixel values of all the partial regions of all the model images input from the learning database 1 and outputs the basis vectors to the shape information database, and converts all the partial regions into the model image local vectors. The feature space generation unit 21 projects the image into the feature space as a model. The feature extraction unit 3 calculates the average and variance of the image local vector and outputs the average and variance together with the shape identifier. The feature vector extraction unit 3 receives the base vector of the feature space from the feature space generation unit 20, and outputs the model image local A shape information database in which the mean and variance of the vector are input and stored together with the shape identifier, 4 is an image input unit for inputting an image whose shape is to be determined, and 5 is a shape identifier stored in the shape information database 3. An image cutout unit which cuts out an image having the same size as the partial shape to be detected from the input image by inputting, 6 a shape classifying means for judging whether or not the image to be detected is present in the image input from the image cutout unit 5; Is characterized as a partial image vector using the basis vector input from the shape information database 3 with the partial image input from the image cutout unit 5. The feature space projecting unit 61 projecting between the object space and the object space to be detected is obtained by calculating the distance between the partial image vector input from the feature space projecting unit 60 and each average of the model image local vectors input from the shape information database 3. A partial shape classification unit for determining whether or not the partial shape matches the shape; and 62, when the partial shape classification unit 61 detects the partial shape of the detection target object, the overall shape of the partial shape detected for each of the detected partial images of the input. The whole shape region estimating unit 63 for estimating the region of the whole object in the input image from the position with respect to the position of the whole object is estimated by the whole shape estimating unit 62. And 7 is a tallying unit.
An output unit that determines that the object is located at a position where the number of times estimated as the position of the object in Step 3 is equal to or greater than a certain value, and displays the position and the detected shape on a display or the like.

The operation of the image recognition apparatus configured as described above will be described below with reference to the flowcharts of FIGS. FIG. 14 is a flowchart showing the operation of the model generation means, FIG. 15 is a flowchart showing the operation from input of image data to be recognized to output of a recognition result, and FIG. 16 is stored in the image database 1. 17 shows an example of a model image, FIG. 17 shows an example of a region to be cut out by the image cutout unit 5 with respect to an image input from the image input unit 4, and FIG. 18 shows an example of a tally result output by the tally unit 63 8 shows an example in which the result detected by the shape classification means 6 is output to the display by the output unit 7. FIG.

Before performing recognition, data of shape information used at the time of recognition is constructed in advance. In the image database 1,
A gray image file of the object as shown in FIG. 16 is stored as a model image. Each image is divided into rectangular local areas of the same size, and for each local area, a shape identifier describing the shape name of the part, the image file name, the upper left corner coordinates and the lower right corner coordinates of the local area is attached. ing. The shape name is composed of the shape name “sedan a” of the entire object and a number indicating which part of the sedan a local area is, and the same number indicates the same position regardless of the image. There may be overlap between the local regions. FIG. 16 is an example of a model image when detecting a sedan having a constant angle with respect to the camera at which the image was taken and a distance from the camera. In actuality, many sedan images having the same appearance are partially identical. Each is stored with a shape identifier.

When a sedan a as shown in FIG. 16 is to be detected, the model generating means 2 extracts from the image database 1 all model images whose shape names are sedan a together with the shape identifier, and generates a feature space. Input to the section 20. The feature space generation unit 20 obtains eigenvalues and eigenvectors from the pixel values of all the local model images, using the local rectangular area indicated by the target area of each shape identifier as a local model image (step 1401). At this time, the local model images are all the same size, and in FIG.
22 pixels. Also, when calculating the eigenvectors, first, create a vector in which pixel values are arranged in a line for each local model image, and then create a vector obtained by subtracting the average vector of all these vectors from each vector, Deriving eigenvalues and eigenvectors.

The feature space generator 20 stores eigenvectors corresponding to the n largest eigenvalues in the shape information database as base vectors (step 1402).
All the local model images are projected onto the eigenspace using the n eigenvectors, and are used as local model image vectors (step 1403). The feature amount extraction unit 21 receives the local model image vector and the shape identifier from the feature space generation unit 20, and calculates the average and the covariance of the local model image vectors having the same shape identifier (Step 1404). The feature amount extraction unit 21 stores the average of all the local model images, the average of the local model image vectors for each shape, the covariance, and the shape identifier in the shape information database 3 (step 14).
05).

When performing recognition, first, image data to be recognized is input from the image input unit 4 (step 15).
01). The image cutout unit 5 includes the shape information database 3
Then, the size of the local model image is calculated from the target area of the shape identifier stored in, and a partial image of the same size as shown in FIG. 17 is sequentially cut out from the input image while moving an arbitrary pixel (step 1502).

The shape classifying means 2 inputs one partial image from the image cut-out unit 5 and
, And the feature space projection unit 60 calculates
Projects a partial image into the eigenspace as a partial image vector
(Step 1503). The partial shape classifying unit 61 receives the partial image vector from the feature space projecting unit 60 and the average vector and covariance of each part of the sedan a from the shape information database 3, and inputs the partial image vector and each average vector. The Mahalanobis distance is calculated (step 150
4) Output the shape identifier of the partial shape having the average vector for which the minimum distance has been calculated.

The whole shape area estimating section 62 outputs the coordinates of the difference between the upper left corner coordinates described in the target area of the shape identifier and the upper left corner coordinates of the input partial image.
Votes for the coordinates (step 1505). Here, the coordinates for voting indicate the position of the object in the input image.

When the feature space projection unit 60 to the tallying unit 63 perform steps 1503 to 1505 for all partial images cut out from the input image (step 1).
506), the tallying unit 63 obtains a tallying result as shown in FIG. In FIG. 18, the coordinates and the number of votes are displayed in descending order of the number of votes. If the number of votes of each coordinate of the total result is larger than a certain value (step 150)
7) It is determined that there is an object at the position of the coordinates in the input image (step 1508), and the output unit 7 changes the shape of the object to the position of the partial image of the object in the input image as shown in FIG. The resulting figures are superimposed and output (step 1510). In step 1507, if there is no vote larger than a certain value,
It is determined that there is no object in the input image (step 150
9) Output the input image as it is (step 151)
0). This output is output from the output terminal 213 via the I / F unit 208.

[0050]

As described above, according to the present invention, even if the object surface has a different color, the object having the same shape can be detected by the shape feature with a small amount of model data, and a part of the object is concealed. Can be detected with high accuracy. Further, the shape and position of the object in the input image can be output with high accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram of an image recognition device according to a first embodiment of the present invention.

FIG. 2 is a block diagram of a computer-based image recognition device according to the first embodiment of the present invention.

FIG. 3 is a flowchart showing a processing flow of a model generation unit according to the first embodiment of the present invention.

FIG. 4 is a flowchart showing a flow of processing from an image input unit to an output unit according to the first embodiment of the present invention.

FIG. 5 is a diagram showing an example of a model image and a shape identifier stored in an image database according to the first embodiment of the present invention.

FIG. 6 is a diagram showing an example of an average image and a shape identifier of a model stored in a shape information database according to the first embodiment of the present invention;

FIG. 7 is a diagram illustrating an example of a rectangular area cut out by an image cutout unit;

FIG. 8 is a diagram illustrating an example of a detection result output from an output unit.

FIG. 9 is a block diagram of an image recognition device according to a second embodiment of the present invention.

FIG. 10 is a flowchart showing a processing flow of a model generation unit according to the second embodiment of the present invention.

FIG. 11 is a flowchart illustrating a flow of processing from an image input unit to an output unit according to the second embodiment of the present invention.

FIG. 12 is a diagram showing an example of a model image and a shape identifier stored in an image database according to the second embodiment of the present invention.

FIG. 13 is a block diagram of an image recognition device according to a third embodiment of the present invention.

FIG. 14 is a flowchart showing a flow of processing of a model generation unit according to the third embodiment of the present invention.

FIG. 15 is a flowchart showing a flow of processing from an image input unit to an output unit according to the third embodiment of the present invention.

FIG. 16 is a diagram showing an example of a model image and a shape identifier stored in an image database according to the third embodiment of the present invention.

FIG. 17 is a diagram showing an example of a rectangular area cut out by an image cutout unit according to the third embodiment of the present invention;

FIG. 18 is a diagram illustrating an example of a tally output by a tallying unit according to the third embodiment of the present invention.

FIG. 19 is a block diagram showing an example of a conventional image recognition device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image database 2 Model generation means 3 Shape information database 4 Image input part 5 Image cutout part 6 Shape classification means 7 Output part 20 Feature space generation part 21 Feature quantity extraction part 60 Feature space projection part 61 Partial shape classification part 62 Overall shape area Estimating unit 63 Totaling unit 201 Computer 202 CPU 203 Memory 204 Keyboard / display 205 Storage medium unit 206-208 I / F unit 209 CPU bus 210 Camera 211 Image database 212 Shape information database 213 Output terminal

Continued on the front page F term (reference) 5B057 AA06 BA02 CE09 DA02 DA07 DA16 DC09 DC22 DC33 5L096 BA04 CA02 EA35 FA18 FA32 FA33 FA81 HA08 JA11 KA04

Claims (8)

[Claims] An apparatus for detecting an object having a specific shape from an image, an image database in which a shape identifier representing a shape of a detection target and an image of the object having the shape are registered in advance as a model image, Model generating means for previously extracting shape features from the model image; a shape information database for storing the shape features and the shape identifiers in pairs; and an image input unit for inputting an image for detecting an object. And an image cutout unit that cuts out a part of the input image as a partial image, and a shape classification unit that compares the partial image with the shape feature and determines whether or not there is a shape to be detected in the partial image. And an output unit that outputs information indicating the shape of the detected object and information indicating the position of the shape in the input image. 2. The model generating means extracts, for each shape, an average image of the model image belonging to the shape and a variance of each pixel of the model image as a shape feature, and extracts an average image, a variance and a shape for each shape. 2. The image recognition apparatus according to claim 1, wherein the identifier and the identifier are stored in a pair in the shape information database. 3. An apparatus for detecting an object having a specific shape from an image, comprising: a shape identifier representing a shape to be detected and an image database in which an image of the object having the shape is registered in advance as a model image; A basis vector of a feature space is calculated from a pixel value of the model image, all the model images are projected as a model image vector in the feature space, and each model image vector to which the shape identifier is added, and a shape is calculated from the model image vector. A model generating means for calculating a characteristic statistic as a shape feature parameter for each of them, and assigning the shape identifier; and a shape information database for registering the base vector and the shape feature parameter from the model generating means together with the shape identifier. And an image input unit for inputting an image for detecting an object, and cutting out a part of the input image as a partial image An image cutout unit, the partial image is projected onto the feature space using the basis vector to obtain a partial image vector, and the partial image vector is compared with a model using the shape feature parameter to be included in the partial image. A shape classification means for determining whether or not there is a shape to be detected; and, if there is a shape in the input image that matches the shape to be detected, information representing the shape and a position of the shape in the input image. An image recognition device comprising: an output unit that outputs information. 4. The method according to claim 1, wherein the calculation of the shape feature parameter for each shape by the model generating means includes calculating the shape feature parameter for each shape from the average vector and the covariance of the model image vector calculated from the model image of the object having the same shape. The image recognition device according to claim 3, wherein the calculation is performed. 5. The model generating means calculates an average image of the model images for each shape, calculates a base vector from pixel values of all the average images, and stores all the model images in a feature space in a feature space. The image recognition apparatus according to claim 3, wherein the image is projected as a vector, and the shape identifier is assigned to each model image vector. 6. The image recognition apparatus according to claim 3, wherein the shape identifier includes information indicating which part of the object is the shape. 7. The shape classification unit estimates at least one region of the entire object in the input image for each shape identifier of the determined partial shape, and totals the estimated region of the entire object for each input partial image. 7. The image recognition apparatus according to claim 6, wherein the position of the object area is output by the operation. 8. A method for detecting an object having a specific shape from an image, comprising: a shape identifier representing a shape to be detected and an image database in which an image of the object having the shape is registered in advance as a model image; Extracting a shape feature from the model image in advance, inputting an image for detecting an object, a shape information database storing the shape feature and the shape identifier in advance, Cutting out the part as a partial image, collating the partial image with the shape feature, determining whether the detection target has a shape in the partial image, and determining the detected detection target shape. And outputting information representing a position of the shape in the input image.