JPH061504B2 - Feature extraction circuit - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character extraction circuit for characters, which is required in a character identification device such as an optical character reader (OCR).

[Conventional technology]

FIG. 6 is an explanatory diagram for explaining a conventional example of a feature quantity extraction method. That is, the character pattern P to be identified
About character frame W and sub character frames W1 to W10 are set,
Area A1 to the character pattern P existing in each sub-character frame
A10 is obtained, and these are combined to be the characteristic amount of the character pattern P. The feature amount extracted in this way is compared with a dictionary consisting of a standard area for each sub-character frame defined in advance for each character and its deviation to identify the input character of interest.

[Problems to be solved by the invention]

However, the feature amount extraction method as described above is effective for a character pattern or the like recorded in print on a managed paper, but for a stamp character or the like in which the character line width greatly changes, it is not possible to use each sub character frame. There is a problem that the variation of the area becomes large and it cannot be applied.

Therefore, an object of the present invention is to provide a feature extraction circuit capable of stably extracting a feature amount even if the character line width changes.

[Means for solving problems]

In order to determine one character area of interest from image data obtained by imaging a character pattern string and binarizing it for each pixel and extracting the feature amount, the contour of the character of interest is tracked over one round, and each contour point Contour tracing means for obtaining a coordinate position and a direction code to the next contour point, and a plurality of contour points for counting the number of contour points having a predetermined direction code for each part of the character of interest determined by the position information from the contour tracing means And a counting unit for counting the count outputs, and the count outputs are combined to obtain the feature amount of the target character.

[Action]

Since the feature amount due to the outline of the character pattern is hardly affected by the character line width, a stable feature amount is extracted using this.

〔Example〕

FIG. 1 is a main view which best shows the features of the present invention, and FIG.
FIG. 1 is an overall configuration diagram showing an example of a general optical character reading device including the present invention, FIG. 3 is an explanatory diagram for explaining a direction code, and FIG. 4 is a detailed description of an embodiment of the present invention. FIG. 5 is an explanatory diagram for explaining the direction code set for each sub-character frame in FIG. 4.

First, FIG. 2 will be described. In the figure, DT is an input, for example, a character pattern string recorded on paper.
Reference numeral 1 is a scanner such as a television camera for converting an input DT into an electric signal by photoelectric conversion, and 2 is various operations for the output from the scanner 1 to digitally change the background and character pattern to "0" and "0". This is a binarization circuit for converting into 1 ". 3 indicates the output from the binarization circuit 2 as the input image DT
According to this, it is a memory for storing one page, and is called a page memory here. Reference numeral 4 denotes a one-character cutout circuit that reads out a pattern area for one character of interest from the character pattern string stored in the page memory 3. Note that the method of determining the one-character area is not particularly relevant here, so the description thereof will be omitted. Reference numeral 5 obtains contour point information by performing contour tracing, which is well known for a 1-character pattern cut out by the 1-character cutting circuit 4, and has a preset direction (setting direction) for each sub-character frame (sub-window). A feature extraction circuit that counts the number of contour points and obtains the feature amount of a character pattern. Details of this circuit will be described later. Reference numeral 6 is an identification circuit for identifying the character amount of the character pattern of interest from the feature extraction circuit 5 by collating it with a dictionary showing the characteristic amount of each sub-window for a known (standard) character that has been set and registered in advance. Is. An editing circuit 7 stores and edits the identification result for each character from the identification circuit 6, and outputs it to the outside as needed.

Now, returning to FIG. 1, the feature extraction circuit will be described in detail. In the figure, reference numeral 51 denotes a one-character memory that stores an image of one character of interest and that can be randomly accessed for contour tracking and the like. Reference numeral 52 denotes a contour tracking unit that extracts the contour pixel of interest and the adjacent pixel information from the memory 51, sequentially determines the next contour point in a predetermined direction (for example, counterclockwise direction), and obtains contour point information of the focused character. The contour tracking unit 52 updates the contour tracking control symbol C each time the next contour point is determined. The control signal C includes the coordinate value (x,
y) and its direction code (V). Here, as shown in FIG. 3, the direction code means that the range of 0 to 360 ° is divided into 8 directions at 45 ° intervals, and 0, 1, 2, ...
It is the one with the number 7. FIG. 4 shows an example in which the contour of the character image P of interest called "5" is traced counterclockwise, for example, starting from the ST point.

Now, as shown in FIG. 4, the character frame of the target character image P is divided into four sub windows W1 to W4, and the direction code as shown in FIG. 5 is set for each of them. Here, the direction codes 1, 2, and 3 (S1, S4) are assigned to the subwindows W1 and W4, and the direction codes (S2 and S3) are assigned to the subwindows W2 and W3. ) Are set respectively. Counters 53A and 53 of FIG.
In B, 53C, 53D, sub-windows W1, W2, W
Direction codes S1, S2, S set for 3 and W4
Count the number of contour points with 3, S4. That is,
In this example, the number of contour points having the direction codes 1, 2 and 3 is set in the sub windows W1 and W4, and
In 2 and W3, the number of contour points having the direction codes 5, 6 and 7 is counted respectively. At this time, since the sub-window determination information SW1, SW2, SW3, SW4 is given to each of the counters 53A, 53B, 53C, 53D, the count value can be obtained for each sub-window. .

By extracting the feature amount as described above, it is possible to obtain a stable feature amount against variations in the character line width, variations in the inclination of characters, etc. As a result, it is possible to perform highly accurate character identification.

〔The invention's effect〕

According to the present invention, contour tracing is performed on a two-dimensional character pattern to obtain a contour position and a direction code, and the number of contour points having a set direction code is counted for each partial area of the focused character, and this is calculated. Since it is set as the feature amount of the character pattern of interest, it is not affected by variations in character line width, etc.
It is possible to obtain a stable feature amount, and as a result, there is an advantage that high-precision character identification is possible.

[Brief description of drawings]

FIG. 1 is a main view which best shows the features of the present invention, and FIG.
FIG. 1 is an overall configuration diagram showing an example of a general optical character reading device including the present invention, FIG. 3 is an explanatory diagram for explaining a direction code, and FIG. 4 is a detailed description of an embodiment of the present invention. FIG. 5 is an explanatory view for explaining the direction code set for each sub-character frame in FIG. 4, and FIG. 6 is an explanatory view for explaining a conventional example of the feature amount extraction method. It is a figure. Description of symbols 1 ... Scanner, 2 ... Binarization circuit, 3 ... Page memory, 4 ... 1 character cut-out circuit, 5 ... Feature extraction circuit, 6
...... Identification circuit, 7 ...... Editing circuit, 51 ...... 1 character memory, 52 ...... Contour tracking unit, 53A to 53D ...... Counter, DT ...... Document, W ...... character frame, W 1 to W 10 ...... sub character Frame, P ... Character pattern.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-57-212582 (JP, A) JP-A-57-161985 (JP, A) Technical Report of the Institute of Electronics and Communication Engineers PRL81-93, (SHO-57-02- 19), pp. 85-92

Claims (1)

[Claims] 1. A feature extraction circuit for determining a character region of interest from image data obtained by picking up a character pattern string and binarizing it for each pixel, and extracting the feature amount thereof. Contour tracing means for tracing the coordinate position of each contour point and a direction code to the next contour point, and a predetermined direction code for each part of the character of interest determined by the position information from the contour tracing means. And a plurality of counting means for counting the number of contour points having, and summing the respective count outputs to obtain the feature amount of the target character.