JPH0585060B2 - - Google Patents

Description

[Detailed description of the invention]

[Industrial Application Field] The present invention relates to a recognition method for recognizing patterns such as characters and figures. [Prior Art] One of the effective means for recognizing patterns such as characters and figures is the correlation method. In devices that recognize patterns such as characters and figures using the correlation method, the original two
It is known that better results can be obtained by first performing feature extraction processing and then correlating higher-order features than performing recognition by calculating correlations from the dimensional image itself.From this perspective, For example, Tokukai Akira
Proposals such as those in Publication No. 59-792 have been made. This method has yielded excellent experimental results, which is evidenced by the fact that the Proceedings of the Institute of Electronics and Communication Engineers (D) J67-D, 2
It was published as ``Improvements of the correlation method for character recognition'' in pp. 224-231 of J67-D, 12 (February 1980), pp. 1442-1449. The outline thereof will be explained below with reference to the flowchart of FIG. Here, after inputting the document to be recognized,
The image data is photoelectrically converted (referenced) by a photoelectric conversion means such as a two-dimensional line sensor and stored in a memory, and this data is aligned, vectorized, and converted by a computer or other calculation means (not shown). Normalize the size, etc. (,
), and then repeat the propagation stop process (,
, Reference) Extract high-order features, perform blur processing by distributing them between vector planes including convolution of a two-dimensional Gaussian weight function (see), and calculate similarity with a predetermined dictionary pattern. It also detects the maximum value (see ) and outputs it as the recognition result. The above will be explained in more detail with reference to FIGS. 6 to 10. Furthermore, Fig. 6 is an explanatory diagram showing a unit direction vector, Fig. 7 is a block diagram showing a spatial differential mask, Fig. 8 is a block diagram showing a coupling coefficient mask, and Fig. 9 explains the concept of propagation stop processing. An explanatory diagram, FIG. 10, is an explanatory diagram illustrating propagation stop processing conditions. First, a black and white binarized image is spatially quantized into meshes of a predetermined size, and for each mesh, the boundaries between the black part (stroke component) and the white background component that make up the characters are extracted, and further, This is expressed as a linear combination that minimizes the sum of squares of coefficients using unit vectors g1 to g8 defined at 45 degree intervals as shown in FIG. In order to easily obtain the linear combination coefficients representing this boundary component, the spatial differential masks MA _x , MA _y shown in FIG. 7 and the combination coefficient masks MB ₁ , MB ₂ , MB ₃ , MB ₄ shown in FIG. 8 are used. In other words,
Using the spatial differential masks MA _x and MA _y , the spatial differential value of the boundary is a set of X direction component and Y direction component ( _▽
y ), and non-zero unit vector numbers and their coefficients are obtained using the coupling coefficient masks MB ₁ to _{MB 4} . In this way, the coupling coefficient P is expressed as P (i, j, k) for X coordinate i, Y coordinate j, and unit vector k, and the vector plane P _k with k as a parameter is expressed as P (i, j, k).
Consider (i, j). That is, P _x (i, j) = P (i, j, k). Next, after normalizing the size of P _k (i, j), the direction-specific pattern Q _k (i, j) is first expressed as
The signal is allowed to propagate in a predetermined direction, and then a stop process is performed. This is repeated a predetermined number of times. The propagation process is to move by a unit position in the direction of the vector component. The orthogonal component is literal, but the diagonal component is separated into horizontal and vertical components and propagated for convenience by using grid points on orthogonal coordinates. Therefore, although there were originally components in eight directions, there are apparently 12 directions (intermediate progress) as shown in Table 1 due to the calculation of the stop and propagation processing. The result of the n-th propagation process is Q _k ⁽ⁿ⁾ (i, j)
Then, the algorithm can be expressed as follows. Q _k ⁽ⁿ⁾ (i+λ _k,j +μ _k )=Q _k ^(n-1) (i, j) where Q _k ^(O) (i, j)=Q _k (i, j).
Note that λ _k and μ _k indicate the direction of unit position movement for each vector plane. Table 1 shows the values of λ _k and μ _k . Further, in the propagation process, arbitrary weighting can be applied to orthogonal components and oblique components as variations thereof. By setting the weight of the diagonal component to 0, only the horizontal and vertical components can be used. However, since the diagonal component is separated into horizontal and vertical components and propagated,
This point must be kept in mind when assigning weights.

[Problem that the invention seeks to solve]

However, the results obtained using this method were obtained in the limited experimental range of the educational kanji database, and it is unlikely that it will be put into practical use within the scope of JIS level 2 in fields such as handwritten character recognition and multi-font printed character recognition. In addition to the handwriting deformities of scribes and the variety of fonts, the appearance rate of characters with many strokes increases, making it difficult to separate complicated characters from each other or from complicated characters and kana characters, so there is still room for improvement. It became clear. Therefore, in view of the above-mentioned points, the present invention improves the feature extraction method so that the structure of characters is explicitly expressed and stable features are obtained that are not affected by handwriting deformation or differences in details between fonts. As a result, it is an object of the present invention to provide a recognition method that can easily separate even complex characters and has a high recognition rate. [Means for solving the problem] The order of the conventional simple feature pattern is increased so that the structure of the character can be expressed explicitly as the depth from the outermost part to the innermost part. That is, in the present invention, instead of propagating in one fixed direction, the propagation direction is alternately reversed so that the propagation and stopping processes are executed a predetermined number of times for each mesh, and then the process is terminated leaving only the stopped components. Do it repeatedly. The components stopped at each end are saved as high-order feature patterns, and during recognition, correlation values are determined for each as an independent feature. [Effect] When the preserved feature patterns are arranged in chronological order, the outermost components at that time will flow out of the feature surface each time there is a reversal, so the deeper components will remain until later. . Therefore, due to the above-mentioned terms, characters that are structurally similar, such as "self" and "shiro", "heaven" and "dai", etc., are well separated, and the recognition rate is improved. [Example] FIG. 1 is a flowchart showing an example of the present invention. This corresponds to Fig. 5, but the same parts as before are omitted and different parts (Fig. 5) are omitted.
Only the parts corresponding to steps ~ in the figure are shown. That is, eight directional patterns Q _k (i, j) of line segments constituting a character are first propagated in a predetermined direction (reference), and then a stopping process is performed (reference). Repeat this a predetermined number of times.
After that, instead of continuing to propagate fixedly in one direction, the components that have not stopped are removed from the feature pattern that has been stopped, and then they are propagated in the opposite direction (see), and in the opposite direction. The propagation stop process is repeated a predetermined number of times as a vector component (see). In this way, the characteristic pattern after the stop processing is saved every time the pattern is reversed, and the pattern is repeatedly reversed M times. Note that the number of repetitions of inversion is determined depending on the capacity of the memory, and may be completed after 4 or 5 times, or may be repeated until there are no more patterns. Usually it is 4-8. This will be explained in more detail below. First, the propagation process is basically the same as before,
The difference is that the propagation direction is reversed each time it is reversed.
The horizontal and vertical components are moved in the direction of the vector component, and the diagonal component is separated into horizontal and vertical components and moved by a unit position. Therefore, although it originally had components in eight directions, due to calculation reasons, it appears to have 12 sides as shown in Table 2. After the m-th inversion, the result of the n-th propagation process is Q _k ⁽ⁿ⁾ (i, j)
Then, the algorithm can be expressed as follows. Q _k ⁽ⁿ⁾ (i+ω _n・λ _k , j+ω _n・μ _k ) = Q _k ^(n-1) (i, j) However, ω _n = (−1) ^(m+1) , Q _k ^{(O )} (i, j)=Q _k (i, j). Here, λ _k and μ _k indicate the direction of unit position movement for each vector plane, and ω _n indicates that it is reversed every other time. Table 2 shows the values of λ _k and μ _k .

〔Effect of the invention〕

According to the present invention, conventional feature patterns are extended to higher-order features, and the structure of a character can be expressed explicitly as depth from the outermost part to the innermost part. As a result, the following effects can be expected. Rather than propagating in one fixed direction,
After performing the propagation and stop processing a predetermined number of times for each mesh, the process of leaving only the stopped components and ending the process is repeated by alternately reversing the propagation direction. High-order features can be extracted. The components stopped at each end are saved as high-order feature patterns, and during recognition, correlation values are determined for each as an independent feature. the result,
Recognition can be efficiently performed in stages from general recognition to detailed recognition. If the saved feature patterns are arranged in chronological order, the outermost components at that time will flow out of the feature surface each time there is a reversal, so the deeper components will remain until later. In other words, the simpler the character (the fewer the number of strokes), the more components remain in patterns with a lower inversion order (m), and the difference in the number of lines appears more clearly in the characteristic pattern. Therefore, due to the above, characters that are structurally similar, such as "self" and "shiro", "heaven" and "dai", etc., are well separated, and the recognition rate is improved.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing an embodiment of the present invention, FIG. 2 is an explanatory diagram for explaining the concept of stop processing according to the present invention, and FIG. 3 is an explanation for explaining propagation stop processing conditions according to the present invention. 4 is an explanatory diagram for explaining the feature extraction method according to the present invention, FIG. 5 is a flowchart showing a conventional example of a pattern recognition method based on the correlation method, and FIG. 6 is for explaining a unit direction vector. 7 is a configuration diagram showing a spatial differential mask, FIG. 8 is a configuration diagram showing a coupling coefficient mask, FIG. 9 is an explanatory diagram for explaining the concept of conventional propagation stop processing, and FIG. 10 is an explanatory diagram for explaining conventional propagation stop processing conditions, FIG. 11 is an explanatory diagram for specifically explaining the process of conventional propagation stop processing, and FIG. 12 is a specific example of conventional propagation stop processing results. It is an explanatory diagram for explaining. Code explanation, V ₀₁ , V ₀₂ ... Vector components of interest,
V _A ~ V _E ... Opposing vector component, g ₁ ~ _{g 8} ... Unit direction vector, MA _x , MA _y ... Spatial differential mask,
MB ₁ to MB ₄ ...Coupling coefficient mask.

Claims (1)

[Claims] 1 Divide the screen containing the characters to be recognized into meshes of a predetermined size, extract the boundary component between the strokes that make up the character and its background for each mesh, represent it as a vector with a predetermined size and direction, and based on this, In the character recognition method that performs recognition,
After performing the propagation and stopping processing a predetermined number of times for each mesh, the process of ending the process leaving only the stopped components is repeated by alternately reversing the propagation direction, and the stopped components are extracted at each end as a high-order feature pattern. Characteristic character recognition method.