JPH01321589A - Feature extraction system - Google Patents

Abstract

PURPOSE:To improve the recognizing performance of characters by deciding a partial area for each direction component to store the information on a character pattern and to extract the direction features. CONSTITUTION:A character pattern memory part 1 stores an input character pattern and a direction extracting part 2 reads the input pattern to extract a distribution pattern of direction components for each direction. A direction pattern memory part 3 stores the distribution patterns for each direction component. Then a direction feature extracting part 4 inputs the distribution patterns for each direction component and obtains the horizontal direction component in each partial area of a lateral rectangle, the vertical direction component in each partial area of a longitudinal rectangle, and the oblique direction component in each partial area of a regular square respectively. Then the part 4 obtains the sum total of the picture element values of distribution patterns of each direction and performs the integration of those patterns to extract and output a direction feature pattern. Thus the character recognizing performance is improved.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to character recognition, particularly to a feature extraction method in handwritten character recognition.

(Prior Art and its Issues) With the diversification of information processing systems, various data input methods are required, and character recognition technology is also being put into practical use as a powerful data input method. However, current character recognition technology is far inferior to Iruma in terms of character reading performance, and a character reading device with higher character recognition performance is desired.

In order to improve character recognition performance, preprocessing, feature extraction processing, and classification/identification processing that constitute character recognition are performed.

Efforts are being made to improve each post-processing process.

Among these, regarding the feature extraction process, the effectiveness of directional features is shown in, for example, the literature Transactions of the Institute of Electronics, Information and Communication Engineers (D), V
ol, J65-D, 隘5. p1), 550-557° Saito, Yamada, Yamamoto "Analysis of handwritten kanji using directional pattern matching method" etc. The conventional method for extracting directional features is to first find the distribution of directional components on the character pattern plane, then divide the character pattern into multiple subregions, and calculate the directionality within each subregion. Integrate ingredients. By processing the character pattern in this way, it is possible to extract a direction feature pattern in which the distribution of direction components is compressed. However, it is difficult to define area divisions during this process, and if the divisions are fine, the amount of data will be large, and the problem will be that the area will be sensitive to noise, including deformation of character patterns, etc.; , the problem arises that information for classifying and identifying characters is lost, making it difficult to identify similar characters.

(Means for Solving the Problems) According to the present invention, in optical character recognition that automatically reads a character pattern given as an M×N grid array, there is provided a method for extracting features of the character pattern. T, M×N
A character pattern storage unit stores a character pattern, and pixels included in a small area of the character pattern are sequentially read from the character pattern storage unit, and the direction on the character pattern is determined by assigning a direction component to a pixel representing the small area. a direction extraction section that creates a direction pattern indicating the distribution of components; a direction pattern storage section that stores the direction pattern created by the direction extraction section; a direction pattern storage section that reads the direction pattern from the direction pattern storage section; , the directional feature pattern for character recognition is performed by integrating within a vertically long area, integrating horizontal components within a horizontally long area, and integrating diagonal components within an area of the same size vertically and horizontally. a directional feature extraction unit that calculates
Character recognition performance can be improved by realizing a feature extraction method including a direction feature storage section that stores the direction feature pattern output from the direction feature extraction section.

(Operation) The principle of the present invention will be explained below. Figure 2 (a) shows the character pattern and character pattern area representing the kanji "sumi", and Figures 2 (b) and (C) each show the character pattern and character pattern area representing the character "sumi".
This is the distribution pattern of the horizontal component and vertical component of "black ink". Conventionally, in order to obtain a directional feature pattern from this distribution pattern, a character pattern area is divided into m x n partial areas, and each partial area is square. However, when there are many horizontal components, such as "black ink", it is desirable to have a high resolution in the vertical direction in order to preserve pattern information.

On the other hand, in the case of characters with many vertical components, it is desirable to have high resolution in the horizontal direction in order to preserve pattern information. Therefore, when integrating horizontal components, the third
As shown in Figure 3(a), each partial area is divided into horizontally long rectangles, and when integration is performed within each partial area and the vertical components are integrated, as shown in Figure 3(b). Next, by dividing each partial area into a vertically long rectangle and performing integration within each partial area, a directional feature pattern that preserves pattern information can be obtained.

In this case, the integration of diagonal components is done as usual in Figure 3 (
As shown in C), division is performed so that each partial area becomes a square, and integration is performed within each partial area. For example, when a character pattern consists of 64 x 64 pixels, to extract the vertical component's directional feature pattern, a partial area is divided into 4 vertically and 16 horizontally. Character recognition It is possible to obtain a directional feature pattern in which pattern information is saved, which is useful for improving recognition performance.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of the configuration of the present invention. The character pattern storage section 1 stores input character patterns, and may be a normal storage means.

The direction extraction unit 2 reads the input character pattern as a signal 11, extracts the distribution pattern of directional components for each direction, and generates a signal 1.
2, the extraction algorithm is shown in the above-mentioned literature, etc., and can be easily realized by conventional technology using ordinary logic elements, storage means, etc. The direction pattern storage section 3 stores the distribution pattern for each direction component sent as the signal 12, and may be a normal storage means.

The directional feature extraction unit 4 inputs the distribution pattern for each direction component as a signal 13, and as explained in the section of the operation, for the horizontal component, it calculates the pixel values of the horizontal distribution pattern in each partial region of a horizontally long rectangle. The vertical component is integrated by finding the sum of the pixel values of the vertical distribution pattern within each subregion of a vertically long rectangle, and the diagonal component is integrated by finding the sum of the pixel values of the vertical distribution pattern in each subregion of a vertically long rectangle. Integration is performed by calculating the sum of the pixel values of the distribution patterns in each diagonal direction within each partial region, and the directional feature pattern is extracted and output as a signal 14.
This can be easily realized using conventional technology using ordinary logic elements, storage means, and the like. The direction feature storage section 5 includes the direction feature extraction section 4.
A normal storage means may be used to store the directional characteristic pattern output as the signal 15 from.

In this example, the extraction of the directional feature pattern was explained by calculating the sum of the pixel values of the distribution pattern within each partial region. It is also possible to normalize the value and use it as a directional feature pattern.

Again, an adder.

Although it is necessary to add a divider, etc., it can be easily realized using conventional technology.

(Effects of the Invention) As described in detail above, according to the present invention, by defining a partial region for each directional component, character pattern information can be preserved and directional feature extraction that is not sensitive to noise can be realized. Therefore, by applying the feature extraction method of the present invention to a character recognition device, character recognition performance can be significantly improved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. FIG. 2 is a diagram illustrating a horizontal component distribution pattern and a vertical component distribution pattern of a character pattern, and FIG. 3 is a diagram illustrating an example of region division performed for extracting directional features in the present invention. In the figure, 1 is a character pattern storage section, 2 is a direction extraction section, 3 is a direction pattern storage section, 4 is a direction feature extraction section, and 5 is a direction feature storage section. Agent Patent Attorney Shinsuke Honjo Figure 1 (a) (b) (c) Figure 2 (a) (b) (c) Figure 3

Claims (1)

[Claims] In optical character recognition that automatically reads a character pattern given as an M×N grid array, a character pattern storage unit stores the M×N character pattern, the system extracting the characteristics of the character pattern. Then, by sequentially reading pixels included in a small area of the character pattern from the character pattern storage unit and assigning a directional component to a pixel representing the small area,
a direction extraction unit that creates a direction pattern indicating the distribution of directional components on the character pattern; a direction pattern storage unit that stores the direction pattern created by the direction extraction unit; Directional components are integrated within a vertically elongated area, horizontal components are integrated within a horizontally elongated area, and diagonal components are integrated within an area of the same size both vertically and horizontally. A feature extraction method comprising: a direction feature extraction section that obtains a direction feature pattern; and a direction feature storage section that stores the direction feature pattern output from the direction feature extraction section.