JPS6048582A - Character cutting-out method of character recognizer - Google Patents

Abstract

PURPOSE:To improve character cutting-out accuracy by dividing again a divided observation pattern in the direction of a character writing row in response to the projection distribution of each divided observation pattern and therefore cutting-out a character by division and redivision. CONSTITUTION:The projection distribution obtained by a projection distribution calculating circuit 2 is fed to a cutting-out line detecting circuit 4. Then the areas having smaller value than the threshold value stored in a register 3 are detected to deliver the 1st character cutting-out line to divide an observation pattern in the character writing row direction. The output of the circuit 4 is supplied to a cutting-out line detecting circuit 6 via a mean value calculating circuit 5 to deliver the 2nd character cutting-out line to divide again the divided observation pattern in the character writing row direction. In addition, the cutting-out lines to be deleted and added are decided by the character cutting-out line detecting circuits 8 and 9. The outputs of these circuits 4, 6, 8 and 9 are supplied to a cutting-out circuit 10 to cut-out the observation pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Technical Field of the Invention The present invention relates to a character recognition method, and more particularly to a character segmentation method for a character recognition device when using a free-formant form.

(b) Background of the technology Character recognition devices are usually equipped with a photoelectric conversion scanning device, etc. as observation 01; and observe the character pattern of each character from a line of characters written on a form (character entry sheet). Convert it into an observed pattern represented by quantized data, or observe a character pattern for one line written on a form and convert it into an observed pattern represented by a quantized signal, and then convert it into an observed pattern represented by a quantized signal. Extract the observed pattern.

- Recognize observation patterns character by character. Observing the character pattern for each character or extracting the observed pattern for each character is called extraction or segmentation, and in order to facilitate this extraction, until now,
A space was provided for each character on the second page of the form.

However, recently, forms commonly referred to as free-memory forms have come into use, forms in which only character lines are provided without character frames for specifying the spacing between Zunawara characters.

(C) Prior art and problems In character recognition devices that use free-formano 1 to form, after observing the character pattern for one line written on the form, Recognition is performed by cutting out the corresponding observed pattern, but cutting out the observed pattern is difficult.

Conventionally, this was done in a secondary manner.

First, as shown in Fig. 1, we first calculate the projection distribution of character pattern forming pixels (usually black pixels) with respect to the axis parallel to two character lines (Y axis), that is, the histogram of black pixels with respect to the Y axis. First, temporary character cutting lines as shown by ↓ marks kO, kl, k2, k3...kl are attached to the parts where the value of the projection distribution of 8 black pixels is small, and these temporary character cutting lines are Calculate the average value of the intervals.

Next, each interval of the temporary character cutting line and the average value are compared, and a part of the temporary character cutting line is removed (
The character cutting position was determined by applying k4) or replenishment (mouth 1 and n2) and 2-ζ.

However, if there are many places in the character line where adjacent characters have complicated phase J1, 6. -c; It was extremely difficult to obtain the desired character cutout position of 16.

(d) Object of the Invention An object of the present invention is to improve the precision of character recognition in a character recognition device using a free form character form.

(e) Structure of the invention What is the character extraction method of the character recognition device according to the present invention?

Specify the character entry interval - 1,4'J' Specify the character entry line.A character recognition device that recognizes characters written on the form. - Divide the observation pattern in the direction of the character entry line and control the division view by the division/J
The divided observation pattern is subdivided in the character entry line direction according to the projection distribution of the llll pattern, and the niIkiviewllltl+ pattern is cut out by the division and the redivision.

(f) Examples of the Invention 1 and 2 The gist of the present invention will be specifically explained by examples.

Figure 2 is a diagram showing the configuration of one embodiment of the present invention.
`` - Observing the character line written in the character entry line and obtaining the observation pattern for each character Lr/1i11 part, 2 is the projection of the observation pattern obtained by observation part 1 in 11ii character recorder 1] Projection distribution d1'#-circuit to obtain the distribution.

3 is a register that stores a threshold value used in the first cut-out 7-line detection circuit 4, which will be described later; a first lino drawing line detection circuit for dividing the observed pattern in the character writing direction by detecting the first character cutting line; 54; l first; The portion obtained by dividing the observed pattern in the character entry 41 direction by the first character cutting line obtained by the cutting 7 line detection circuit 4;';
+J observation bar observation bag - an average value calculation circuit that obtains the average value of the projection distribution for each song; 6 is a circuit that calculates a predetermined number smaller than 1 from the projection distribution of the divided observation pattern to the average value of the projection portion 4j for each divided observation pattern In order to redivide the divided observation pattern in the character entry line direction by detecting f1^(, which is smaller than the threshold value obtained by multiplying Cutout line detection circuit,
7 is the average interval of each divided observation pattern cut out by the first cutting line and the second cutting line.
The average value 61 circuit for obtaining α, 8 is the interval of each divided observation pattern cut out by the first cutting line and the second cutting line, and ζ obtained by the average value calculation circuit 7.
Compare with 11 average value.

A third cutting line detection circuit 9 determines a cutting line to be deleted from among the first cutting line and the second cutting line; The interval between the IGII observation patterns to be cut out and the average value calculated by the average value calculation circuit 7 are compared, and the cutting line to be added to the =-th cutting line and the second cutting line is calculated. The fourth cutout line detection circuit 210 to be determined is detected or determined by the first cutout line detection circuit 4, the second cutout line detection circuit, the third cutout line detection circuit 8, and the fourth cutout line detection circuit 9. 11 is a feature extraction circuit that extracts features for each divided observation pattern cut out by the cutting circuit 10; 12 is a shuttle finisher for each character type; 1' is a dictionary memory for recording features 1a; 13 is a recognition circuit that recognizes the divided observation pattern by comparing the features of the divided observation pattern obtained by the feature extraction circuit 11 with the dictionary memo 1/2;

Zunawara 1 First, let's look at the first cutting line detection circuit 4.

As shown by the ↓ mark in Fig. 3 ta+, 1 (0, kl, k2...kl is obtained as the first cutting line that appears at the point where the value of the projection distribution of black pixels is smaller than a predetermined threshold value). It will be done.

Next, in the second cutting line detection circuit 6, as shown in FIG. ml・m2 and m3 are obtained as the second cutting line.At this time.

The second cutting line cannot be obtained between kl and 2 and between 5 and 1 (6).

The reason why the second cutting line cannot be obtained between 2 and 2 and between 5 and 6 is that the average value of the projection distribution for each divided observation pattern obtained by the first cutting line This is by using a threshold value obtained by multiplying by a predetermined number smaller than 1.

After that, in the third cutting line detection circuit 8, as shown in FIG.
As shown in FIG. Lines n1 and n2 are determined.

According to the above embodiment, the cutout line is not only determined using an arbitrary threshold (i#j) but also obtained from the average value of the projection distribution of the 1-segment observation pattern fυ by the second cutout line detection circuit 6. After the cutting line is set using a threshold value, the cutting line is deleted or added using the average value of the interval between each divided observation pattern, so that a uniform character cutting position can be obtained.

(g) As described in the detailed description of the invention, the present invention improves the character recognition accuracy of the method using a free form.
It is possible to improve the extraction accuracy and improve the character recognition accuracy.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of a conventional example, FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention, and FIG. 3 is an explanatory diagram of the embodiment. In the figure, 1 is an observation unit, 2 is a projection distribution calculation circuit, 4 is a first slope J outline detection circuit, and 5 is an average value calculation circuit. 6 is a second cutting line detection circuit I, and 10 is a cutting circuit. 2 towns! Juchi Tae 2 Crab, J! , l #P f prefectural correction route Kirimoto Ritake t 1'''''7y Experiment, 1:, Circuit-1; yo) Roy m,. gTch huge: [RoF7honshigiK Dojir@≦C5 paddle 111j Road average I1 1 shot 2? , q M'3: * , * JL H sticky j4 shame moe lunch soba, 1. j // /θ Shosoft Furnace Job Circuit ri'lct'Bk,, upper,,. Suiryouuta (α) To' 2/ To

Claims (2)

[Claims] (1) In a character recognition device that recognizes characters written on a form that specifies only the character entry line without specifying the character entry interval, the character recognition device recognizes the characters written in the character entry line of the form.
ii) Divide the observation pattern in the direction of the character entry line using the projection distribution for the character entry line of the view/1tll pattern obtained by dividing the divided observation pattern using the projection distribution of the divided view Δ111 Bakun Jun obtained by the division. A character cutting method for a character recognition device, characterized in that the pattern is subdivided in the character writing line direction, and the observed pattern is cut out by the dividing and the re-dividing. (2) The character recognition device according to claim 1, wherein the subdivision of the divided observation pattern is performed using a threshold value obtained from the average value of the projection distribution for each divided observation pattern. Cutting method.