JPH0324709B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention provides highly practical characters that can stably and efficiently extract features for broad classification identification by reading a large number of characters including kanji. Relating to a reading device.

[Technical background of the invention]

Conventional character reading devices extract features of an input character pattern, match them with features of dictionary patterns registered in advance in a dictionary, and read and recognize the input character pattern according to the matching results. is being carried out. However, when a large number of characters including Chinese characters are to be read in this character recognition, the amount of processing required to match the above-mentioned features becomes extremely large, and the processing efficiency deteriorates significantly. Conventionally, prior to the above recognition process, characters to be recognized are roughly classified based on a number of general characteristics, and this reduces the number of characters to be matched at the identification stage, improving processing efficiency and increasing processing speed. Efforts are being made to make this happen. At this time, it is necessary to use the same features as the features used for the above-mentioned major classification and the features used for candidate character identification, but this feature is necessary to prevent deformation of character patterns caused by handwritten characters and various noises. It needs to be sufficiently stable.

However, conventionally, the features used to broadly classify character patterns have focused on the complexity of the character lines, the shape of the peripheral part of the character, and even the rough shape of the entire character. There are things that have been done. In addition, these are introduced in detail in, for example, the following literature.

Sakai, Watanabe “Current status of printed kanji recognition” Information Processing, Vol. 22 No. 4 PP. 274-279 (April 1981) Mori, Sakai “Development of printed kanji OCR that reads 2000 character types at 100 characters/second” Nikkei Electronics, October 31, 1977 issue
PP.102-128 Such conventional feature extraction focuses on the shape of the periphery of the original character pattern, as shown in Fig. 1, for example, and scans rectangular scanning areas U, D, Detect character lines existing in L and R,
The amount is quantized into three levels of "0", "1", and "2" and used as feature data (four-sided code method). In addition, the one shown in Figure 2 focuses on circumscribed sides u, d, l, and r that touch the original character pattern, calculates the area of the character background that touches these circumscribed sides u, d, l, and r, and calculates this by It is used as a multidimensional feature vector (peripheral pattern method). The difference is that the former (Figure 1) focuses on the character line area, while the latter (Figure 2) focuses on the area of the background area, but both are characterized by the area quantity around the character pattern. It can be said that this is true.

In addition, focusing on the complexity of the entire character pattern, the length of the character line is divided into vertical and horizontal directions, and these values are normalized by the character size to calculate the average character line density (complex There is also a known method for calculating the index (Special Publication No. 53-29416).

[Problems with background technology]

However, these feature extraction methods are very effective when the pattern is standardized, such as printed characters, but when the pattern is standardized, such as handwritten characters, they are significantly deformed even though they have the characteristics of characters. In such cases, it is extremely unstable. Furthermore, even if the characters are printed, if the character pattern is chipped or faded, feature extraction becomes extremely unstable. In other words, there is a problem in that feature extraction cannot always be performed stably against deformation of character patterns and various noises.

In order to solve these problems, the present inventors previously published Japanese Patent Application No. 56-208306, Japanese Patent Application No. 57-46749, etc.
A new feature extraction method is proposed. However, a new problem remains to be solved in terms of how to easily extract features that are mutually independent and complementary in behavior with respect to noise.

[Purpose of the invention]

The present invention was made in consideration of these circumstances, and its purpose is to read a large number of characters, including handwritten kanji, and to identify the major classifications of input original character patterns. It is an object of the present invention to provide a highly practical character reading device that can extract features stably and efficiently.

[Summary of the invention]

The present invention detects a circumscribing frame consisting of circumscribing sides in four directions (top, bottom, right, left, and right) of an original character pattern that has been binary quantized and stored, and divides the area surrounded by the circumscribed frame into two in the vertical and horizontal directions, respectively. The upper half,
The lengths of the character lines of the original character pattern in each of the lower half, left half, and right half partial regions are determined, and the partial complexity of the original character pattern is calculated from the relative values of the lengths of these character lines with respect to the circumscribing frame. The character pattern is extracted as feature information and the original character pattern is broadly classified and identified.

At the same time, in addition to the above partial complexity information,
Information on the center of gravity position or area of the character background portion in each of the partial regions is obtained, or information on the average unevenness amount obtained from the relative value of the length of the character contour line of the character background portion with respect to the circumscribing frame is comprehensively used. The original character pattern is roughly classified and identified.

〔Effect of the invention〕

Thus, according to the present invention, the character lines of the original character pattern in each of the upper half, lower half, left half, and right half partial regions formed by dividing the region surrounded by the circumscribing frame of the original character pattern into two vertically and horizontally. The partial complexity of the above-mentioned original character pattern is determined from the relative value of the length with respect to the circumscribed frame, and this is used as information for major classification identification. The feature information can be obtained with sufficient stability against noise. Therefore, it is possible to stably and efficiently perform broad classification identification of input patterns, and it is possible to significantly improve character recognition processing.
In addition, even when dealing with complex character patterns such as kanji, handwritten characters, and even printed characters, the above features can be achieved using the simple method described below. Since it can be easily obtained, it can have a highly practical and remarkable effect in character recognition processing.

Furthermore, in addition to the partial complexity of each partial region mentioned above, information on the center of gravity position and area of the character background, and information on the amount of local unevenness are used to perform general classification identification of the original character pattern. Since features of the original character pattern that are independent and complementary in behavior to noise can be easily and accurately extracted, it is possible to accurately and stably identify the major classifications. Therefore, great practical effects can be achieved, such as being able to accurately recognize and read and input a wide variety of characters with various deformations.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings.

FIGS. 3a to 3d show feature extraction processing of the average number of lines, which is the partial complexity of the original character pattern, in the apparatus of the embodiment, and here a handwritten kanji pattern for "shi" is shown. This character pattern is subjected to photoelectric conversion, then binary quantized, and stored in a pattern storage device. For the original character pattern given in this way, the feature extraction process first begins with the circumscribed edges u in four directions surrounding the original character pattern, up, down, left and right,
A circumscribing frame consisting of d, l, and r is detected. The above circumscribed edges u, d, l, and r are determined by scanning and detecting the character pattern and comparing the position coordinates of the character lines with each other. The right circumscribed edge r touches the right edge, and the upper circumscribed edge u touches the top edge.
and the lower circumscribed side d touching the bottom end. The size of this circumscribing frame is determined as the distance V between the upper and lower circumscribing sides u and d, and the distance H between the left and right exterior sides l and r.

After that, the area surrounded by the circumscribed frame is divided into two in the vertical direction to define an upper half partial area U and a lower half partial area D, and the above area is similarly divided into two in the left and right direction. A left half partial region L and a right half partial region R are respectively defined. Then, the lengths of the character lines of the character pattern in each of the partial areas U, D, L, and R are determined. The length of this character line can be determined by dividing the local area of the middle character pattern in Figure 3a into the diagonal, vertical and horizontal directions as indicated by circles, and Figures 3b to d show the length of the character line. As shown, from the pixel data array of a micro area consisting of 2 × 2 pixels, the unit length is ls, lt, l, ly.
It is obtained by counting the number of occurrences in each direction. That is, the character line length L L in the left half partial area _L is determined by scanning the character pattern in the same partial area L and sequentially inspecting the patterns in the minute area, as L' _L = 1/2 {Σlx + Σls/2 + Σlt}. Desired. Similarly, for the right half partial region R, L' _R = 1/2 {Σlx + Σls/2 + Σlt}, and for the upper and lower half partial regions U and D, L' _U = 1/2 {Σly + Σls/2 + Σlt} It is obtained as L' _D = 1/2 {Σly + Σls/2 + Σlt}. However, these character line lengths
The relative values of L _L , L _R , L _U , and L _D with respect to each of the circumscribed sides l, r, u, and d are determined by using the size of the circumscribed frame as a normalization coefficient, L _L =L' _L /H/2= 1/H{Σlx+Σls/2+Σlt} L _R =1/H{Σlx+Σls/2+Σlt} L _U =L' _U /V/2=1/V{Σly+Σls/2+Σlt} L _D =1/V{Σly+Σls/2+Σlt} Each is required. Physically, the character line lengths L _L , L R , L U , and L _D in each partial area L, _R , _U , and D obtained in this way are horizontal line components in each of the left and right half areas. It represents the average number of lines, that is, how many vertical line components are present in each region of the upper half and the lower half. In other words, this information indicates the local complexity of the character pattern in each partial area.

In this way, by following information on the local complexity of each region, it is possible to obtain characteristic information such as the bias, sides, and crowns unique to kanji, which can be used for broad classification identification as a preliminary step to recognition processing. It becomes possible to use the information effectively as information. In particular, if such information is used, the original character pattern input by hand can be
Since it is possible to easily and accurately recognize whether a Chinese character consists of ``bia'' and ``side'', a kanji consisting of a crown and a foot, or something else, its practical advantage is very high.

By the way, in this device, in addition to the information on the local complexity in the partial area mentioned above, information on the center of gravity position (area) of the character background part in each partial area,
Information on the average amount of unevenness obtained from the character contours is also used to broadly classify input character patterns.

For example, the center of gravity of the character background is each circumscribed edge u,
Using d, l, and r as reference positions (reference lines), scan each area U, D, L, and R from each circumscribed side u, d, l, and r in the direction toward the center of the character pattern to reach the character line. It is calculated by finding the scanning line length up to that point. However, from the viewpoint of signal processing, it is preferable that the maximum length of the scanning is defined as, for example, a length l (variable) depending on the size of the character pattern. That is,
Each circumscribed edge u, which is represented by the sum of the scanning line lengths,
The area S of the background portion of the character pattern corresponding to d, l, and r is determined. For example, the area S of the background portion on the circumscribed side l is determined by scanning in the direction toward the center of the character pattern with the circumscribed side l as a reference position using a scanning line defined to have a maximum length H/2 as shown in FIG. Find its recurrence. However, when the scanning line reaches the character line, the scanning ends, and if the scanning line does not reach the character line, the scanning ends when the length l is reached. Then, assuming that each point of the area indicated by the area S obtained in this way is (x, y), the center of gravity position G (X, Y) of the area S is
Find it as follows.

X=∬ _s x・S _(x

Claims (1)

[Scope of Claims] 1. Means for photoelectrically converting and inputting the original character pattern, means for binary quantizing and storing the input original character pattern, and four directions of the stored original character pattern, top, bottom, left and right. means for detecting a circumscribed frame consisting of circumscribed sides; means for determining the length of each character line of the original character pattern; means for detecting the partial complexity of the original character pattern from the relative value of the length of these character lines with respect to the circumscribing frame; A character recognition device comprising means for broadly classifying and identifying the original character pattern according to the information of the original character pattern. 2 The length of the character line in each partial area is 2×2
2. The character reading device according to claim 1, wherein the length of the boundary line of the original character pattern in a minute region of pixels is determined by integrating the lengths of the boundary lines for each direction of the boundary line. 3. A means for photoelectrically converting and inputting the original character pattern, a means for binary quantizing and storing the inputted original character pattern, and a circumscription consisting of the circumscribed sides of the stored original character pattern in four directions: top, bottom, left, and right. means for detecting a frame, and character lines of the original character pattern in each of the upper half, lower half, left half, and right half partial regions obtained by dividing the region surrounded by the circumscribed frame into two in the vertical direction and the horizontal direction, respectively. means for determining the length of each of the character lines, means for detecting the partial complexity of the original character pattern from the relative value of the length of these character lines with respect to the circumscribed frame, and the position of the center of gravity of the character background in each of the partial areas. or means for determining the relative value of the area value with respect to the circumscribing frame, and means for detecting the character outline of the character background portion in each of the partial areas, and determining the relative value of the character outline length with respect to the circumscribing frame for the original character. means for determining an average amount of unevenness around the pattern; and information on at least one of the average amount of unevenness, the center of gravity position of the character background part, or its area, and the information on the partial complexity to form the original character pattern. 1. A character reading device comprising: means for identifying major categories.