JP3381803B2 - Tilt angle detector - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a tilt angle detecting device. 2. Description of the Related Art In the process of automatically extracting and recognizing character information contained in an image, a process of detecting a line in which characters are arranged in a line or an inclination of an individual character requires high precision reading. Is the key to doing so. However, in recent years, the technology of character recognition has been widely applied to various uses, and it has become necessary to handle even images in which background noise is mixed and characters are unclear. In such a situation, it is strongly expected that not only the character recognition but also the inclination angle detection processing which assists the character recognition is robust against the deterioration of the image quality. Conventionally, a method has been proposed in which a white run length histogram of a divided area is obtained, and an inclination angle is obtained from an offset between areas obtained by a cross-correlation between adjacent histograms (Japanese Patent Laid-Open No. 4-177584).
issue). According to this method, the processing time is short, and it is possible to detect even if figures, tables, photographs, and the like are mixed in the document image. Therefore, there is a problem that the inclination angle cannot be accurately detected. A method has also been proposed in which the sum of the absolute values of the differences in the number of black pixels between adjacent lines in a plurality of directions is calculated, and the direction in which this is the maximum is defined as the direction of inclination of the document (Japanese Patent Laid-Open No. 3-18888). issue). In this method, when there is noise between character lines, or when the character is thin and partially missing, the change in the sum of the black pixel numbers hardly appears in the scanning direction, and the detection is unstable. Problem. [0005] A method has also been proposed in which a histogram of black pixels is created for each of a plurality of slopes, and the histogram is evaluated based on the sum of portions exceeding a predetermined threshold, and the angle at which the maximum is obtained is defined as the slope angle. (JP-A-3-83188)
issue). According to this method, a binary image of a document is scanned along a plurality of predetermined gradients as shown in FIG. 14, and a histogram of black pixels is created for each gradient. Then, as shown in FIG. 15, the portions exceeding the predetermined threshold value of the black pixel frequency are summed, and the histogram having the largest value is processed as the inclination of the document. According to this method, it is considered that the inclination of the document can be correctly detected if the threshold value is appropriately set even when the document has a background stain. However, if the threshold value is too high or too low, a correct inclination cannot be obtained. A table in which the threshold value of the histogram of the black pixels is determined in advance from the paper size is prepared. However, even if the paper size is the same, the height of the histogram varies depending on the character capacity, print density, etc. fluctuate.
Therefore, it is difficult to appropriately determine the threshold value of the histogram of the black pixels in advance. In other words, it has been difficult to correctly determine the tilt angle when the size and number of characters and the state of printing change in various ways. [0007] As described above, a conventional histogram of black pixels is created for each of a plurality of gradients, and the histogram is evaluated by the sum of portions exceeding a predetermined threshold value. In the method in which the inclination angle is set to the inclination angle, the unevenness clearly appears in the histogram created just at the inclination angle. Therefore, if an appropriate threshold value is provided, the sum of the portions exceeding the threshold value should be the largest. However, if the threshold value is not appropriate, this evaluation value does not always match the unevenness of the mountain. In addition, an appropriate threshold greatly depends on noise mixing and the size and type of characters, and there is a problem that it cannot be set in advance only by the size of an image. The present invention has been made in view of the above circumstances, and provides a method for stably detecting a line / character inclination angle even for an image in which background noise is mixed or characters are unclear. Aim. In the tilt angle detecting apparatus according to the present invention, a histogram is formed by scanning a read image at a plurality of predicted tilt angles and generating a black pixel cumulative histogram for each tilt angle. Evaluation value calculating means for adding a difference between a maximum value and a minimum value of each section of the black pixel cumulative histogram over the entire range of the black pixel cumulative histogram to obtain an evaluation value at each tilt angle; The evaluation amounts at the corners are compared, and a tilt angle determination unit that uses the value of the tilt angle at which the evaluation amount takes the maximum value as the tilt angle of the image is used. [0010] The value of the evaluation amount is the sum of the difference between the maximum value and the minimum value in the section width L of the black pixel cumulative histogram scanned at the inclination angle θ over the entire range of the histogram. And the printing state are less affected by the tilt angle θ
Can be detected with high accuracy. Here, the parameter determined in advance is the section width L, where L is an amount determined according to the size of the main character in the image or the width of the line, and is one of the character size or the line width. It is desirable to set the value to about 1/2 to 1 time. Embodiments of the present invention will be described below in detail with reference to the drawings. In this method, a bookstore mark as shown in FIG. 1 is read. A read image of a bookstore stamp as shown in FIG. 1A is binarized as shown in FIG. 1B, and a part of FIG. 1A is extracted as shown in FIG. 2A. A histogram is created from the binary image as shown in FIG. 2 (b), and a tilt angle is detected using the histogram. A histogram is created for each of a plurality of predicted tilt angles, and this histogram is created. From the above, the difference between the maximum value and the minimum value within the interval is used as the evaluation amount, and the value obtained from the sum over the entire range of the histogram is used as the evaluation amount, and the value of θ that takes the maximum value is used as the inclination angle. And FIG. 3 shows a flowchart of the tilt angle detecting step. The evaluation value C _θ is obtained for a plurality of inclination angles set in advance (step 101).
When this process is completed, _{θ at} which _Cθ becomes the maximum is set as the optimum tilt angle (step 108), and the detection of the tilt angle ends. When detecting the evaluation amount C _θ , first, the scanning is performed in the direction of the inclination angle θ to obtain the black pixel cumulative histogram H.
_θ (x) is obtained (step 102). [0018] then x _i is set to an initial value x _s, the evaluation value C _theta is set to 0 (step 103). Then, the maximum value and the minimum value of the black pixel cumulative histogram H _θ (x) are obtained in the section [x _i , x _i + L] (step 104). Thereafter, the evaluation value C _θ is obtained by adding the difference between the maximum value and the minimum value of the black pixel cumulative histogram H _θ (x) to the initial value (= 0) of the evaluation value C _θ (step 10).
5). Thereafter, x _i = x _i +1 is updated (step 106). The steps 104 to 106 are defined as x _i + L
Is repeated until the other end x _e (Step 1
07). In this embodiment, since the direction of the thin ruled line does not always coincide with the inclination of the bookstore mark, the histogram of the histogram corresponding to the ruled line is usually used to prevent the detection of the histogram portion corresponding to this ruled line. After the portion is removed by the minimum value filter (indicated by a broken line as step 200), the evaluation amount is calculated (steps after step 103). Generally, step 200 is not particularly necessary. FIG. 4 shows the result. The left side is a histogram not subjected to the minimum value filter, the right side is a histogram subjected to the minimum value filter, and the numerical value on the right side is the sum of the difference between the maximum value and the minimum value. Here, the portion surrounded by the thick frame takes the maximum value, and this tilt angle is set as the tilt angle of this bookstore mark. In this way, the evaluation value C _θ is obtained for a plurality of tilt angles, and _{θ at} which C _θ becomes the maximum is determined as the optimum tilt angle. When the inclination angle is detected in this way, bookstore mark recognition is performed along this inclination angle. FIG. 5 shows a flowchart of this process. That is, after detecting the inclination angle (step 20)
1), a line to be recognized is cut out (step 20)
2) After that, individual characters are cut out (step 20).
3). FIG. 6 shows the result of line segmentation. Then, the character is determined from the cutout, and the character is determined (step 204). Here, the width L between the maximum value and the minimum value of the histogram is preferably about 1/2 to 1 times the line interval W. In this manner, the inclination angle θ can be detected with high accuracy without being affected by the number of characters or the state of printing. As a specific example, FIG.
7B, the inclination angle is detected and the line is cut out as shown in FIG. 7B. As a result, the line can be cut out as shown in FIG. 7C. Next, a method for detecting the inclination angle of a character will be described as a second embodiment of the present invention. In this method, as shown in FIG. 8, a cut-out and recognition process is performed by detecting the inclination angle of a character string written in italics. FIG. 1 is a simple flowchart of a method for correcting a handwritten character as shown in FIGS. 9A and 9B, for example.
0 is shown. FIG. 9 (a) shows the original binary image shown in FIG. 9 (b) with its inclination corrected. First, a histogram in an oblique direction is created at intervals of Δθ (step 301). The histogram is represented as H _θ (x) as a function of x as shown in FIG. And the evaluation quantity C of the following formula
_θ that maximizes θ (here, 60 degrees) is set as the tilt angle of the character (step 302). [0036] Thereafter, the image is rotated in the θ direction and the upright processing ends (step 303). In this way, handwritten characters can be cut out and recognized well. FIG. 12 shows a flowchart of this recognition processing. In this method, a handwritten numeral image is input (step 401), and is binarized (step 402).
L is set to about the average character width, and the above-described tilt angle detection processing is performed (step 403). The read image is corrected by the tilt angle, and the pixel values of the coordinates on the scanning line are shifted as shown in FIG. Create Write Correction Image (Step 4
04). Thereafter, segmentation recognition of the character string is performed (step 405). As described above, according to the present invention, the inclination angle θ can be detected with high accuracy without being affected by the number of characters or the state of printing.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a read image of a bookstore mark and a binarized image thereof according to an embodiment of the present invention. FIG. 2 is a partially extracted image of the binarized image and a histogram thereof. FIG. FIG. 4 is a flowchart of a tilt angle detection process using the tilt angle detection device according to the embodiment; FIG. 4 is a histogram of black pixels at each tilt angle using the device according to the embodiment of the present invention (the left side is not subjected to the minimum value filter; FIG. 5 shows the sum of the difference between the maximum value and the minimum value of the histogram. FIG. 5 is a flowchart showing a bookstore mark recognition process using the apparatus of the embodiment of the present invention. FIG. 6 is a diagram showing a line cutout result. FIG. 7 is a diagram showing a bookstore mark recognition process using the apparatus of the embodiment of the present invention. Figure 9: Original binary image used in this method FIG. 10 is a view showing an image and a binary image after the correction. FIG. 10 is a flowchart showing the same detection processing. FIG. 11 is a histogram of black pixels at each inclination angle in the same processing. FIG. 12 shows the sum of differences between the maximum value and the minimum value of the histogram. FIG. 12 shows a handwritten character recognition process using the handwritten character inclination angle detection method according to the embodiment of the present invention. Flowchart FIG. 13 is an explanatory diagram of creating a tilt-corrected image by the same method. FIG. 14 is a diagram showing detection of a tilt angle using a conventional method. FIG. 15 is a diagram showing detection of a tilt angle using a conventional method.

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-3-83188 (JP, A) JP-A-1-312682 (JP, A) JP-A-1-305488 (JP, A) JP-A-2- 69886 (JP, A) JP-A-57-52971 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1/00 G06K 9/18- 9/44 G06K 9/62-9/82 H04N 1/40

Claims (1)

(57) [Claim 1] Histogram creating means for scanning a read image at a plurality of predicted inclination angles and creating a black pixel cumulative histogram for each inclination angle; An evaluation amount calculating unit that adds a difference between a maximum value and a minimum value for each section over the entire range of the black pixel cumulative histogram to obtain an evaluation amount at each inclination angle, and compares the evaluation amount at each inclination angle. And a tilt angle determination unit that sets a value of the tilt angle at which the evaluation amount takes the maximum value as a tilt angle of the image.