JP2010067068A - Image feature classifying apparatus and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately classify image features of a target image, by narrowing the elements which are characteristic to the image features from a feature quantities used, when learning classification. <P>SOLUTION: An image forming apparatus includes: a classifier 252 having learned the classification of image features, on the basis of sample images prepared for each classification of the image features; a feature quantity extracting part 210 for extracting a plurality of feature quantities from the target image to generate a feature vector; a variation information acquiring part 254 for acquiring the variation of each of the plurality of feature quantities in the classification for each classification of the image features; a feature quantity selecting part 256 for selecting a partial feature quantity from the plurality of feature quantities, on the basis of the variation of each of the plurality of feature quantities obtained for the classification for each classification of the image features; a distance calculating part 258 for calculating the distance between the feature vector, in a feature space represented by the selected partial feature quantity and each classification of the image features; and a classification determining part 260 for determining the classification of image features possessed by the target image, on the basis of the result of comparing the calculated distances. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image feature classification device and a program.

An image formed on paper may be scanned and defects included in the image may be classified based on image characteristics of the formed image. At this time, as described in Patent Document 1 below, various feature amounts are extracted from the scanned image to generate a feature vector, and the distance between the previously learned classification and the feature vector for each image feature is calculated. Based on this, there is a technique for determining the classification of image features of a scanned image.
JP 2001-134763 A

  Depending on the classification of image features, the features that make up the learned feature space include features that cannot be said to be the features of the classification. Using these features for classification can lead to misclassification. Sometimes the classification could not be narrowed down and the classification could not be performed accurately.

  One of the objects of the present invention is to provide an image feature classifying apparatus and program for classifying image features of a target image with high accuracy by narrowing down features characteristic to the image features from the feature quantities used during classification learning. It is in.

  To achieve the above object, the invention of the image feature classification apparatus according to claim 1 is based on a feature vector having a plurality of feature amounts obtained from a specimen image prepared for each classification of image features as elements. Classification learning means for learning the classification of the image features; generation means for acquiring the plurality of feature amounts from the target image; and generating a feature vector having the acquired feature amounts as elements; and the image features Acquisition means for acquiring the variation of each of the plurality of feature amounts in the classification for each of the classifications, and variation of the plurality of feature amounts acquired by the acquisition means for the classification for each classification of the image features. Based on the selection means for selecting a part of the feature quantity from the plurality of feature quantities, and the generation means in the feature space represented based on the part of the feature quantity selected by the selection means. And calculating a distance between the generated feature vector and each classification of the image feature, and determining a classification of the image feature of the target image based on a comparison result of the distance calculated by the calculation unit. Classification determination means.

  According to a second aspect of the present invention, in the image feature classification apparatus according to the first aspect, the selection unit selects the feature amount in ascending order of variation from the plurality of feature amounts.

  According to a third aspect of the present invention, in the image feature classification apparatus according to the first or second aspect, the calculation unit calculates a Mahalanobis distance between the feature vector generated by the generation unit and each classification of the image feature. The classification determining unit calculates and classifies the image feature classification of the target image as an image feature classification corresponding to the minimum Mahalanobis distance calculated by the calculation unit.

  According to a fourth aspect of the present invention, the image feature classification is learned based on a feature vector whose elements are a plurality of feature amounts acquired from a sample image prepared for each image feature classification. A classification learning unit, a generation unit that obtains the plurality of feature amounts from the target image, generates a feature vector having the acquired feature amount as an element, and the image feature in the classification for each classification of the image features. An obtaining unit that obtains a variation of each of the plurality of feature amounts; and for each classification of the image features, based on the variation of each of the plurality of feature amounts obtained by the obtaining unit for the classification, from the plurality of feature amounts A selection means for selecting a part of the feature quantity; and a feature vector generated by the generation means in a feature space represented based on the part of the feature quantity selected by the selection means; A computer serving as a classification determining unit for determining a classification of an image feature of the target image based on a comparison result of the distance calculated by the calculation unit and a distance calculated by the calculation unit. It is made to function.

  According to the first aspect of the present invention, it is possible to classify the image features of the target image with high accuracy by narrowing down some of the feature amounts from the feature amounts used during classification learning.

  According to the second aspect of the present invention, characteristic information can be selectively used for classification of each image feature.

  According to the third aspect of the present invention, it is possible to determine the classification of the image features of the target image by comparing the calculated distances in common and comparing the calculated distances.

  According to the fourth aspect of the present invention, it is possible to cause the computer to function so as to classify the image features of the target image with high accuracy by narrowing down some of the feature amounts used in the classification learning.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments (hereinafter referred to as embodiments) for carrying out the invention will be described with reference to the drawings.

  FIG. 1 shows a functional block diagram of an image forming apparatus 10 having a failure diagnosis function according to the present embodiment. As shown in FIG. 1, the image forming unit 100 includes an image reading unit 102, a sensor unit 104, a diagnostic information input unit 106, and a failure diagnosis unit 108. In the embodiment, the configuration of the image feature classification apparatus according to the present invention is applied to a defect type determination unit 212 provided in the image forming apparatus 10 described later.

  The image forming unit 100 has a function of forming and outputting an image on printing paper based on image data. The image forming unit 100 converts, for example, image data into raster data, forms a latent image on the photosensitive member with laser light based on the converted raster data, and transfers the toner image to the printing paper after attaching the toner to the photosensitive member. Then, the image may be formed by a laser printer method for forming an image. In addition, the image forming unit 100 may form an image by an electrophotographic method, a thermal method, a thermal transfer method, an ink jet method, or the like.

  The image reading unit 102 has a function of optically reading a document to be read and obtaining a scanned image. The image reading unit 102 is an original placing table such as a platen glass on which an original is placed, an optical system that optically reads the original placed on the original placing table, and an image read by the optical system is converted into image data. And an image processing unit.

  The sensor unit 104 includes a plurality of sensors, and acquires state information of the image forming apparatus 10 including an operation state of components of the image forming apparatus 10, an internal environment state, a consumable use state, and the like.

  The diagnosis information input unit 106 includes an operation panel and has a function of receiving input of information used for failure diagnosis. The diagnosis information input unit 106 may accept information input such as operation input from a user and start of failure diagnosis, for example.

  The failure diagnosis unit 108 is a function that performs failure diagnosis of the image forming apparatus 10 based on information input from the image reading unit 102, the sensor unit 104, and the diagnosis information input unit 106. FIG. 2 is a functional block diagram showing details of the failure diagnosis unit 108.

  As shown in FIG. 2, the failure diagnosis unit 108 includes a component state information acquisition unit 200, an internal environment information acquisition unit 202, a consumable material information acquisition unit 204, a history information management unit 206, an image defect detection unit 208, and a feature amount extraction. Unit 210, defect type determination unit 212, additional operation information acquisition unit 214, failure cause estimation unit 216, and diagnosis result notification unit 218.

  The component state information acquisition unit 200 is a function that acquires component state information indicating the state of the components constituting the image forming apparatus 10 based on the information acquired by the sensor unit 104. The component state information acquisition unit 200 acquires, for example, information such as paper passage time, drive current value, applied voltage value to the photosensitive member, vibration, operation sound, and light amount as the component state information.

  The internal environment information acquisition unit 202 is a function that acquires internal environment information indicating the state of the internal environment of the image forming apparatus 10 based on the information acquired by the sensor unit 104. The internal environment information acquisition unit 202 acquires, as internal environment information, information such as device internal temperature, device internal humidity, paper temperature, and paper humidity, for example.

  The consumable material information acquisition unit 204 has a function of acquiring consumable material information indicating the state of the consumable material of the image forming apparatus 10 based on the information acquired by the sensor unit 104. The consumable material information acquisition unit 204 acquires, as the consumable material information, for example, information such as the remaining amount of various printing sheets and the remaining amount of toner of each color.

  The history information management unit 206 is a function for managing the usage history of the image forming apparatus 10. The history information management unit 206 is acquired by, for example, a print job processing history, a counter value indicating the number of printed sheets for each component, the component state information acquisition unit 200, the internal environment information acquisition unit 202, and the consumable material information acquisition unit 204. Various types of information may be managed as history information.

  The image defect detection unit 208 has a function of detecting an image defect from the inspection image read by the image reading unit 102. The image defect detection unit 208 detects an image defect from the inspection image based on a predetermined threshold value for the inspection image. For example, if the inspection image is a white-based test chart, a gradation value corresponding to the background color is set as a threshold value, and if the inspection image has pixels with gradation values smaller than the threshold value, It may be detected that there is an image defect.

  When the image defect detection unit 208 detects that there is an image defect, the feature amount extraction unit 210 extracts a plurality of feature amounts from the image area corresponding to each defect.

  The defect type determination unit 212 has a function of determining the type of defect occurring in the acquired inspection image based on the feature amount extracted by the feature amount extraction unit 210. The defect type determination unit 212 generates, in advance, a cluster of image feature amounts representing the types of defects based on the characteristic images in which the defects have occurred, and extracts feature amounts. The cluster to which the feature quantity extracted by the unit 210 belongs may be classified based on the relationship between the feature quantity and the representative feature quantity of each cluster. As one aspect of the classification, the feature vectors may be classified into clusters having the shortest distance from the cluster representative vector. Note that the defect type may be determined using the Mahalanobis distance using the distribution information of the feature amount group for each defect type. Details of the configuration of the defect type determination unit 212 and the processing performed in the defect type determination unit 212 will be described later.

  Depending on the failure location of the image forming apparatus 10, the image defect may not be reproduced in the inspection image. For example, when the failure part is a part of the print engine, the defect is reproduced in the inspection image, but when the failure part is a part of the image reading unit 102, the image defect may not be reproduced in the inspection image. In this case, if the inspection image is set in the image reading unit 102 and the inspection image is read, a defect appears in the scanned image. The additional operation information acquisition unit 214 has a function of acquiring information on an additional operation performed by changing the operation condition of the image forming apparatus 10 in order to specify the cause of such an image defect. The information acquired by the additional operation information acquisition unit 214 is input to the failure cause estimation unit 216.

  The failure cause estimation unit 216 includes a reasoning unit 216A and a failure candidate extraction unit 216B, and has a function of estimating a failure cause based on information input from each unit. For example, the inference unit 216A may be configured by an inference engine that calculates a probability (failure cause probability) that each cause candidate causing a failure is a main cause of the failure that has occurred based on each acquired information. Moreover, the failure candidate extraction unit 216B extracts failure cause candidates based on the failure cause probability calculated by the inference unit 216A. For example, the failure candidate extraction unit 216B may extract failure causes whose calculated probabilities are equal to or higher than a preset value, or extract failure causes up to a preset order in descending order of the calculated probabilities. It is good as well.

  Here, a Bayesian network may be used as an inference engine for calculating the failure cause probability. The Bayesian network is a network in which causal relationships between a plurality of events are sequentially connected and expressed as a network having a graph structure, and dependency relationships between events are expressed by a directed graph. A node is created for each event, and the occurrence probability is assigned to the node as a variable. In addition to the Bayesian network, other methods such as an expert system and a neural network may be used for the inference engine.

  When using the Bayesian network for the inference engine, the inference unit 216A inputs the acquired various information to the Bayesian network and calculates the probability value of each node. The failure candidate extraction unit 216B then identifies the main cause node based on the calculated probability value and extracts it as a failure cause candidate. Thus, based on the failure cause candidates extracted by the failure cause estimation unit 216, failure determination results (failure presence / absence, failure location, failure content), failure prediction results (failure presence / absence, failure location, failure content) ), Or the contents of the examination, the acquired operation state signal, and the like are output to the diagnosis result notification unit 218.

  The diagnosis result notifying unit 218 is a function for notifying a user or the like of the diagnosis result input from the failure cause estimating unit 216. The notification of the diagnosis result may be performed by displaying it on a display included in the operation panel of the image forming apparatus 10 or by displaying it on a management computer connected to the image forming apparatus 10. The diagnosis result notification unit 218 may transmit the diagnosis result to, for example, a management server connected via a network, or may output the diagnosis result to the administrator by outputting it from the image forming apparatus 10.

  FIG. 3 shows a detailed configuration of the defect type determination unit 212. As illustrated in FIG. 3, the defect type determination unit 212 includes a sample information storage unit 250, a classifier 252, a variation information acquisition unit 254, a feature amount selection unit 256, a distance calculation unit 258, and a classification determination unit 260. In this embodiment, the defect type is used as one form of the image feature classification, but the image feature classification is not limited to this.

  The specimen information storage unit 250 stores specimen information of an image corresponding to the type of defect for each type of defect. The sample information of the image may be the image data itself, or may be feature amount information such as a feature vector extracted from the image.

  The classifier 252 is a classifier that has learned the type of defect based on the sample information of the image for each type of defect stored in the sample information storage unit 250. Specifically, the classifier 252 uses, for example, a feature amount vector composed of N image feature amounts extracted from the image by the feature amount extraction unit 210 to calculate the distribution of each defect type in the N-dimensional feature space. Hold. Then, the classifier 252 determines the type of defect that classifies the image to be processed based on the distance between the feature vector obtained from the image to be processed and the type of defect in the feature space.

In FIG. 4, when there are three feature amounts, each defect type G _i (i = 1, 2, 3, ₃ ) in the feature space formed by the _three feature amounts (X ₁ , X ₂ , X ₃ ). ..)) And the feature vector T obtained from the image to be processed. The classifier 252 calculates the distance between the representative vector of each defect type G _i and the feature vector T obtained from the image to be processed, and classifies the image to be processed into the type of defect having the smallest distance. .

  However, if the distance between the feature vector obtained from the processing target image and each defect type exceeds a predetermined threshold, the classifier 252 does not determine the defect type for classifying the processing target image. . In the present embodiment, when the defect type (classification) is not determined by the classifier 252 as described above, a part of the feature quantities is selected from N feature quantities for each defect type, The type of defect is re-determined based on the selected feature amount. Hereinafter, functional blocks related to the defect type redetermination process will be described.

  For each defect type learned by the classifier 252, the variation information acquisition unit 254 represents a variation for each of the N feature amounts constituting the feature vector based on the sample information of the image belonging to the defect type. Get the index value. In this embodiment, the standard deviation is used as the index value of the variation, but the present invention is not limited to this, and other index values such as variance, range, quartile range, average difference, and average absolute deviation may be used. Absent.

FIG. 5 shows an example of a variation information table that stores the standard deviation of each feature quantity acquired for each type of defect by the variation information acquisition unit 254. In the variation information table shown in FIG. 5, the type of defect, that is, classification G _i (i = 1, 2,..., M) and feature amount X _j (j = 1, 2,..., N). The standard deviation acquired for each combination with is stored. In FIG. 5, σ _ij represents the standard deviation of the feature quantity X _j in the classification G _i .

  For each defect type, the feature amount selection unit 256 selects the feature amount in ascending order of the standard deviation acquired by the variation information acquisition unit 254 for the defect type. The number of feature amounts selected by the feature amount selection unit 256 may be determined in advance.

FIG. 6 shows an example of a selected feature amount table showing the feature amounts selected for each defect type. In the selected feature amount table shown in FIG. 6, XG _i indicates a set of feature amounts selected for the defect type G _i , and elements of XG _i have the first to pths in ascending order of standard deviation. The feature amount of the position is selected.

  The distance calculation unit 258 calculates the distance between the feature vector to be classified and the type of each defect in the feature space constituted by the set of feature values selected for each type of defect. The distance calculated by the distance calculation unit 258 may be the Mahalanobis distance.

FIG. 7 shows an example of a distance information table in which information on distances calculated by the distance calculation unit 258 is stored. D _kl in FIG. 7 represents the distance between the feature vector to be classified and the defect type G ₁ in the feature space constituted by the feature amount set XG _k selected for the defect type G _k . Note that k = 1, 2,..., M, l = 1, 2,.

The classification determination unit 260 determines the type of defect that classifies the feature vector to be classified based on the distance calculated by the distance calculation unit 258. Specifically, the classification determination unit 260 determines to classify the feature vector to be classified into the defect type G ₁ corresponding to the minimum distance among the distances included in the distance information table described above. In this way, the classification is determined by the existing classifier 252 by selectively using feature quantities having small variations for each defect type, in other words, by using a feature space suitable for expressing the defect type. Classification is also determined for feature vectors of objects that have not been performed.

  Next, a series of failure diagnosis processing performed by the image forming apparatus 10 according to the present embodiment will be described with reference to the flowcharts shown in FIGS.

  FIG. 8 is a flowchart showing the overall flow of the failure diagnosis process. First, the image forming apparatus 10 starts a failure diagnosis mode based on an operation by a user, and prints out a test pattern for failure diagnosis (S301). Here, the test pattern to be printed out may be held in advance in the image forming unit 100 shown in FIG. The image forming apparatus 10 reads the printed test pattern by the image reading unit 102 (S302).

  Next, the image forming apparatus 10 compares the image (scanned image) read by the image defect detecting unit 208 with a reference image stored in advance, and detects an image defect from the scanned image (S303). If no image defect is detected by the above processing (S304: N), the image forming apparatus 10 has detected that the defect that occurred before that is accidental or the cause of the image defect before the test pattern is output. Is displayed on the operation panel, and the process is terminated. On the other hand, when an image defect is detected by the above process (S304: Y), the image forming apparatus 10 extracts a feature amount from the image area including the detected image defect (S305).

  The image forming apparatus 10 determines the type of defect included in the scan image using the extracted feature amount (S306). Details of the defect type determination process will be described later.

  The image forming apparatus 10 includes various information necessary for failure diagnosis, such as status information of each component constituting the apparatus, history information such as a counter value indicating the number of printed sheets for each component, and environmental information such as temperature and humidity inside the apparatus. Get the data. Further, the image forming apparatus 10 infers the cause of the failure based on the extracted feature amount of each defect, the determined defect type, and the acquired various data (S307). Failure reason inference may be performed by inputting the above various data into a Bayesian network as an inference engine and calculating the occurrence probability of each failure cause.

  The image forming apparatus 10 extracts a failure cause candidate having a high probability of causing a failure based on the calculated occurrence probability (S308). Note that the number of failure cause candidates to be extracted may be determined in advance, or may be specified by inputting an arbitrary number by the user. The image forming apparatus 10 displays the extracted failure cause on the operation panel and notifies the user (S309). If there is no additional operation information at this stage (S310: N), that is, if failure cause candidates can be narrowed down, the diagnosis process is temporarily terminated, and if there is additional operation information at this stage (S310: Y). ), The image forming apparatus 10 changes the operation condition according to the additional operation information and re-outputs the test pattern.

  Based on the re-output test pattern, the user inputs additional operation information to the image forming apparatus 10 (S311). The additional operation at this time is to examine whether or not the defect occurrence state has changed, and may be, for example, enlargement / reduction of an image, output of a test pattern held at each part of an image path, or the like. Therefore, the additional operation information is easily input to the operation panel by the user. Then, the image forming apparatus 10 recalculates the failure cause probability based on the added information and the already input information, and narrows down failure candidates from the calculation result. If failure candidates are narrowed down or there is no information to be added (S310: N), the diagnosis process is temporarily terminated. If there is an image area to be the next diagnosis target (S312: Y), the process returns to S306 to perform failure diagnosis processing for the image area corresponding to the defect, and if there is no image area to be the next diagnosis target ( S312: N), the process is terminated.

  FIG. 9 is a flowchart of the defect type determination process. First, the image forming apparatus 10 generates a feature vector of a scanned image (classification target image) based on the feature amount extracted in S304 (S401).

  The image forming apparatus 10 calculates a distance from the generated feature vector for each defect type (image feature classification) learned in advance (S402). In calculating the distance, the image forming apparatus 10 extracts feature amounts from a plurality of sample images prepared for each defect type, and calculates the centroid of a feature vector composed of the extracted feature amounts. The center of gravity of the calculated feature vector is set as a representative vector. Then, the image forming apparatus 10 calculates the distance between the feature vector generated for the classification target image and the representative vector of each defect type. The above distance may be the Mahalanobis distance obtained by standardizing the Euclidean distance in the feature space by the distribution of the feature vector of the sample image for each defect type.

Next, the image forming apparatus 10 determines whether or not there is a calculated distance between the feature vector to be classified and each defect type that is equal to or smaller than a predetermined threshold (S403). When the image forming apparatus 10 determines that there is no defect type with a distance equal to or smaller than the threshold (S403: N), the image forming apparatus 10 selects one of the unprocessed defect types G _i (i = 1 to M) ( S404) Referring to the variation information table storing the variation of the value for each feature amount for the defect type, p pieces (p <N) of N feature quantities constituting the feature vector in ascending order of standard deviation. ) Is selected (S405). The selection number p of feature quantities may be determined in advance based on the result of evaluation in advance.

The image forming apparatus 10 calculates the distance between each defect type G _i and the feature vector in a feature space represented by p pieces of feature quantities selected (S406). By selectively utilizing the characteristic feature amount of defect type G _i, the actual correlation of the distance between the feature vector and the defect type is reflected.

  The image forming apparatus 10 determines whether or not there is an unprocessed defect type (S407). If there is an unprocessed defect type (S407: Y), the process returns to S404 and the subsequent processing is repeated. . On the other hand, when determining that there is no unprocessed defect type (S407: N), the image forming apparatus 10 classifies the target image into the defect type corresponding to the minimum distance among the calculated distances (S408). ). In S403, when the image forming apparatus 10 determines that there is a defect type with a distance equal to or smaller than the threshold (S403: Y), the target image is classified into the defect type corresponding to the smallest distance among the calculated distances. (S408). The above is the defect type determination process performed by the image forming apparatus 10.

  Of course, the present invention is not limited to the above embodiment. For example, in the above embodiment, the number of feature amounts selected by the feature amount selection unit 256 is determined in advance, but the number of feature amounts to be selected for each image feature is changed as follows. May be. Here, FIG. 10 shows an example of a graph in which the distance between the number of feature quantities selected for a certain image feature classification and the target feature vector is calculated. The feature quantity selection unit 256 uses the number (p) of feature quantities when the feature vector distance is minimum in the graph shown in FIG. 10 for this image feature. The above processing is performed for each image feature. For the number of feature quantities selected for this image feature, a numerical range with an upper limit, a lower limit, etc. may be set.

  Note that the functions of each unit constituting the image forming apparatus 10 according to the present embodiment are such that a program stored in a computer-readable information storage medium is read into the image forming apparatus 10 which is a computer system using a medium reading apparatus (not shown). It may be realized by being executed. This program may be supplied to the image forming apparatus 10 by an information storage medium, or may be supplied via a data communication network such as the Internet. Further, a program executed by the central processing unit and various data may be held in a storage device including a memory element such as a RAM and a ROM, a hard disk, and the like.

1 is a functional block diagram of an image forming apparatus according to an embodiment. It is a detailed functional block diagram of a failure diagnosis unit. It is a detailed functional block diagram of a defect kind determination part. It is a figure which shows the relationship between the classification | category of each image feature in a feature space, and a feature vector. It is a figure which shows an example of the dispersion | variation information table. It is a figure which shows an example of the selection feature-value table. It is a figure which shows an example of a distance information table. It is a flowchart of a failure diagnosis process. It is a flowchart of a defect kind determination process. Shown is an example of a graph that calculates the distance between the number of selected feature values and the target feature vector

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Image forming apparatus, 100 Image forming part, 102 Image reading part, 104 Sensor part, 106 Diagnosis information input part, 108 Fault diagnosis part, 200 Component state information acquisition part, 202 Internal environment information acquisition part, 204 Consumable material information acquisition part 206 history information management unit 208 image defect detection unit 210 feature amount extraction unit 212 defect type determination unit 214 additional operation information acquisition unit 216 failure cause estimation unit 216A inference unit 216B failure candidate extraction unit 218 diagnosis Result notification unit, 250 specimen information storage unit, 252 classifier, 254 variation information acquisition unit, 256 feature quantity selection unit, 258 distance calculation unit, 260 classification determination unit.

Claims (4)

Classification learning means for learning the classification of the image features based on a feature vector having a plurality of feature amounts acquired from a sample image prepared for each classification of the image features,
Generating means for acquiring the plurality of feature amounts from the target image and generating a feature vector having the acquired feature amounts as elements;
Obtaining means for obtaining a variation of each of the plurality of feature amounts in the classification for each classification of the image features;
Selection means for selecting a part of the feature quantities from the plurality of feature quantities based on variations of the plurality of feature quantities obtained by the obtaining means for the classification for each classification of the image features;
Calculating means for calculating a distance between the feature vector generated by the generating means and each classification of the image features in a feature space represented based on a part of the feature amount selected by the selecting means;
A classification determination unit that determines a classification of image features of the target image based on a comparison result of distances calculated by the calculation unit;
An image feature classifying device comprising: The image feature classification apparatus according to claim 1, wherein the selection unit selects a feature amount in ascending order of variation from the plurality of feature amounts. The calculating means calculates a Mahalanobis distance between the feature vector generated by the generating means and each classification of the image features;
The classification determination unit determines the classification of the image feature of the target image as the classification of the image feature corresponding to the minimum Mahalanobis distance calculated by the calculation unit. The image feature classification apparatus described. Classification learning means for learning the classification of the image features based on a feature vector having a plurality of feature amounts acquired from a sample image prepared for each classification of the image features,
Generating means for acquiring the plurality of feature amounts from the target image, and generating a feature vector having the acquired feature amounts as elements;
Obtaining means for obtaining a variation of each of the plurality of feature amounts in the classification for each classification of the image features;
Selection means for selecting a part of feature quantities from the plurality of feature quantities based on variations of the plurality of feature quantities obtained by the obtaining means for the classification for each classification of the image features;
Calculating means for calculating a distance between the feature vector generated by the generating means and each classification of the image features in a feature space represented based on a part of the feature amount selected by the selecting means;
A program that causes a computer to function as a classification determination unit that determines a classification of image features of the target image based on a distance comparison result calculated by the calculation unit.

Images