JPS62137685A - Recognition device for handwritten character by multi-stage matching - Google Patents

Abstract

PURPOSE:To shorten a recognition processing time for an input unknown pattern by making reference patterns into a multiple layered structure, and performing a multi-stage DP matching process. CONSTITUTION:A handwritten character recognition device consists of a representative reference pattern vector storage part 1, a reference pattern storage part 3 included in an area setting the vector of the part 1 as center of gravity, and DP matching arithmetic parts 2 and 4. Assuming that an unknown input pattern is inputted, and the number of images are the m-number of images, the DP matching process is performed with the representative reference pattern vector of a (j) picture which satisfies (m-beta<=j<=m+alpha) and the DP matching arithmetic part 2. As a result, a representative vector yij having the nearest distance is found and a reference pattern group 3 corresponding to an ij is selected. Next, the DP matching process is performed between the input pattern and the reference pattern as the DP matching arithmetic part 4, and the most assembled reference pattern is outputted as a recognition result.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to handwritten character recognition devices, and more particularly to multiple division of standard patterns.

Conventional technology Conventionally, in this typesetting handwritten character recognition, standard patterns are prepared in advance for each character to be recognized, and when an unknown pattern is input, it is compared with all standard patterns and the most Judgment was made by selecting a similar standard pattern.

Problems to be Solved by the Invention However, in order to perform matching processing on unknown input pattern and all standard patterns, the above-mentioned conventional handwritten character recognition device converts characters to be recognized into JIS 1.
When expanded to this level, the matching process becomes enormous.

Even if the standard patterns to be matched are limited depending on the number of strokes in the input kataan, there are many kanji characters with 11 to 73 strokes, which has the disadvantage that it takes a long time to recognize them.

The present invention has been made in order to eliminate the above-mentioned drawbacks inherent in the conventional technology, and therefore, an object of the present invention is to provide a novel handwritten character recognition method that makes it possible to shorten the recognition processing time for input unknown patterns. The goal is to provide equipment.

Means for Solving the Problems In order to achieve the above object, the handwritten character recognition device according to the present invention is characterized in that standard patterns are multi-layered and multi-stage DP matching processing is performed. , it is possible to shorten the recognition processing time for input unknown patterns.

More specifically, the handwritten character recognition device according to the present invention has a configuration as shown below. That is, the standard pattern of characters to be recognized is described as a time series of stroke direction coefficients and stroke lengths. Now, let's use this standard pattern in general: :l! =(& 1. b 1+ a 2. b 2+
+++ + an, bn), the standard pattern I can be considered as an N-dimensional vector having N=2n components, and the standard pattern I can be expressed as one point in the N-dimensional space.

In the conventional DP matching method, for a written input character with the number of strokes m, DP matching is performed on all standard patterns whose number of strokes satisfies m-β≦≦m+α, and the standard pattern with the smallest distance is selected. was.

In the present invention, a standard pattern vector space with one stroke is divided into clusters, and a representative [4 pattern vector y 1J (J=7.-i, . . . , k)] of each cluster is determined. Now, when unknown pattern 2 is input and the number of strokes of this input pattern is m strokes, each representative standard pattern vector Vl, j (1=
j−β, j−β+! ,...,j+α−/,j+α)(j
=/.

Performs DP matching processing with Y,..., k) to obtain the representative vector y, which is the closest distance. That is, d(z.

Ylj)≦d (z, y, n) (d (x, y)
is the DP matching process of 1 and y). Next, DP matching is performed with the standard pattern vector within the area having this representative standard pattern vector as the center of gravity.

Find the most similar standard pattern and use this as the recognition result. If a large number of standard pattern vectors exist within a region, by further dividing this region, the DP matching process can be reduced, and the recognition processing time can be reduced by %.

Embodiment Next, a preferred embodiment of the present invention will be specifically explained with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. An embodiment of the handwritten character recognition device according to the present invention includes a representative standard pattern vector storage section, a standard pattern storage section including an area whose center of gravity is the representative standard pattern vector storage section, and a DP matching calculation section. Reference number / in FIG. 1 is a representative standard pattern vector storage unit, which is created by a clustering algorithm described later. Now, if unknown input cover turn 2 is input and the number of strokes of this input cover turn is m strokes, then DP matching processing is performed in the DP matching calculation unit Y with a representative standard pattern vector of 5 strokes satisfying m-β≦j≦m+α. . As a result, the closest representative vector ylj is found, and the standard pattern group 3 corresponding to ijK is selected. Next, DP matching processing is performed between the input cover pattern and these standard patterns in a DP matching calculation section, and the most similar standard pattern is taken as the recognition result.

FIG. 1 shows a block configuration for creating a representative standard pattern vector. In FIG. 1, j indicates the entire standard pattern of characters to be recognized, and this standard pattern 5 is classified according to the number of strokes as shown in 6. For each standard pattern of each number of strokes, a clustering algorithm is used to calculate representative standard pattern vectors and their corresponding regions. Here, we use Engineering EEE Transaction on Communications (Engineering EEE TRA) as a domain decomposition method.
NS -AOTION ON COMMUNICATION ONS Vol, COM xg, No, /.

Jannary /910. A clustering algorithm such as LINDE described in PP, Zau-gs) is used.

The application of the LINDE clustering algorithm to standard patterns is outlined below. Clustering algorithm by LINDE, (θ), A = (21, χ2....+ :fn)
A: W quasi-pattern alphabet δ=, 0.00/, ni=θ M=/sha (ha=total(A
J 仝(・) Finds the center of gravity of φ.

(1) Divide the total (good) into two by adding a small vector to the given project (good).

Divide y□ into +ε and y□-8.

C is an infinitesimal constant vector M4-ko x M (=), m = OD 1 = -ω.

(3), d(xj, yi)≦d(
If xJ, y, 0, the optimal region division P (Am (V) = (sl: i = /, 2
Find ゜...M).

The average distance between the obtained y□ (i = /, . . . , M) and the standard pattern Xj (J = /, . . . n) is determined.

Dm=D ((2m (MJ , P (Am (Ml
) )(ri), or Dm + -Dm )/Dm <
If it is 1, go to (6), otherwise go to (yo).

(5), Meeting mHQ= 仝(P(Am(M)))=仝(
1) Return to i=/, -, j-m+m+/ (, y).

(6), all one ← A M If M is the number of divisions of the standard pattern vector space that you want to find, that's it.

Otherwise, return to (1).

As described in detail, according to the present invention, standard patterns are double-layered, and when K is the number of N1 divisions, which is the total number of standard patterns to be recognized, the DP matching process is approximately K + N.
/K.

In addition, the standard pattern is layered into M layers, and the recognition of each layer is completed, so that the processing time can be shortened.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the multi-stage DP matching handwritten character recognition device according to the present invention, and FIG. 2 is a block diagram showing the creation of a representative standard pattern indicated by the reference number / in FIG. l...Representative standard pattern vector storage unit, 4...
・DP matching calculation unit, 3...Standard pattern corresponding to 1j, j...Standard pattern to be recognized, 6...Standard pattern with the same number of strokes Patent applicant NEC Co., Ltd. Agent Patent attorney Yutabe Noya Figure 1 Figure 2

Claims (1)

[Claims] Handwritten character recognition using pattern matching includes means for multiple-dividing a standard pattern using a clustering algorithm, and means for matching an input unknown pattern with a representative standard pattern created by the multiple division means. A multi-stage matching handwritten character recognition device.