JPS5983280A - Pattern recognizing method - Google Patents

Abstract

PURPOSE:To improve the degree of resemblance between frames separated from each other, by defining an average value of the n-th and the (n+1)-th patterns or the (n+1)- th pattern as a standard pattern in accordance with the result of the comparison between a prescribed value and the degree of resemblance which is obtained by comparing both patterns with each other. CONSTITUTION:The first frame A of patterns A-D of a negative film 1 is defined as a standard pattern, and the degree of resemblance between the frame A and the next frame A' is discriminated by calculation to output discrimination signals S1-S5; and if this degree is higher than a prescribed value S3, it is decided that both patterns have a resemblance to each other or equal to each other, and average feature quantity data of frames A and A' is defined as a new standard pattern 2. The frame B and the pattern 2 are compared with each other to attain the degree of resemblance between them, and the pattern of the frame B is stored as a standard pattern 3 if the discrimination signal is prescribed value S2 or lower. The frame A'' is compared with patterns 2 and 3 to calculate the degree of resemblance; and in this case, since the frame A'' has a resemblance to the pattern 2, the averaging processing between the frame A'' and the pattern 2 is performed to generate and sotre a standard pattern 4. Hereafter, patterns of frames C, A''', and D are recognized similarly to generate standard patterns 5-7.

Description

[Detailed description of the invention]

The present invention relates to a pattern recognition method. For example, a single negative film often contains frames of the same or similar scenes, and it is necessary to print the same scene several times. Most of the same scene often differs in detail depending on the position and angle of the camera, the size and position of the main subject, the background, etc. Therefore, the maximum density 1) nax and the minimum density Dmin A large difference occurs between the maximum concentration Dmax and the minimum threshold])
Even if the difference from min is small, a difference occurs when the calculation output is converted into a stepwise conversion. As a result, even the same scene is printed in different dark IWs. Incidentally, the same scene is often photographed consecutively. Therefore, we calculated the total area average transmission density (Large Area 'I') of several frames before and after the frame to be measured.
anceDensity; hereinafter simply referred to as LATD)
There is a determination method that compares the detected value of the shooting light quality and the detected value of the frame, and if the difference is within -Y value, it is determined that these are the same scene (for example, Japanese Patent Publication No. 49-40942). If it is determined that they are the same scene, the same scene frames are printed with the same density using the intermediate value of the exposure amounts of these same scenes. Here, as is clear from Evans' principle, most of the pieces have a constant LATD. Therefore, the method of determining only by the values of LA and TD is as follows:
Even if they are different scenes with different image configurations, they are often determined to be the same scene. L depending on the exposure control force method
Even if A'l'l) is the same, the density will be changed and printed due to the influence of other characteristic values. Furthermore, if it is determined that the scenes are the same because the LA'I'D is the same as described above, a problem arises in that even if the scenes are different, they will be printed with the same density as the previous and subsequent frames. For this reason, conventionally, adjacent frames of a negative film 1 are compared as shown in FIG. 1, and similar patterns are recognized from the degree of similarity. In other words, frame A is compared with the adjacent frame A', and if they are similar, frame A' is treated as the same or similar pattern, and the next frame A' and the immediately adjacent frame A' are compared. Compare with
If these are similar to each other, frame A'' is determined to be the same or similar scene.In this way, it is determined whether or not they are the same scene based on the degree of similarity between adjacent frames. Therefore, for adjacent frames, the judgment of the same scene is accurate, but for example, 2 frames 1l11c, A and A''' are dissimilar even though they are dissimilar. , there is a drawback that they become similar. Therefore, the object of the present invention is to provide a method of recognition that overcomes the above-mentioned drawbacks. This invention will be explained below. This invention is based on a method of recognizing consecutively arranged patterns by one degree of similarity, and n
The (1≦n≦N)th pattern and the (n+1)th pattern are compared to determine the degree of similarity, and if the degree of similarity is smaller than a predetermined value or below, the (n+1)th pattern is is stored as a 4M quasi-pattern, and when the degree of similarity is greater than or equal to a predetermined value, the n-th and (
Standard patterns are obtained and stored by averaging processing of the (n+1)th pattern, and these stored standard patterns are sequentially compared with other patterns, and a new standard pattern is created based on the degree of similarity. It was designed so that it could be formed. That is, to explain the case where the patterns A to I) of the negative film 1 as shown in FIG.
The degree of similarity is determined by subtracting the degree of similarity described later, for example, in five stages 81 , 82 , 83 , 84
, 85 identification signals are output. If the identification signals 81 to S5 are greater than or equal to a predetermined value (for example, 83), it is determined that they are similar or identical, and the average feature amount data (described later) of frames A and A' is set as the new standard pattern 2. . Note that this averaging process may be an arithmetic average of the patterns of two frames (or an average value weighted according to the number of frames).Then, the next frame B and this newly set standard pattern may be 2, and if the identification signal is 82 or less, the pattern of this frame B is stored as is as another standard pattern 3. Next, the next frame A" is The degree of similarity is calculated by comparing them with standard patterns 2 and 3. In this case, since frame A'' is similar to standard pattern 2, average processing of frame A'' and standard pattern 2 is performed to create a new pattern.・Form and memorize standard pattern 4.Furthermore, the next frame C and the moment pattern 2 found so far
-4 are compared to calculate the similarity of each, but in this example, frame C has identification number 46 of 82 or less for any standard pattern 2-4.

Therefore, the pattern of this frame C is stored as a new standard pattern 5. Thereafter, pattern ba for frames A, "', and D is performed in the same manner, or in this example, a new standard pattern 6.
7)1. This will be configuration 1. In this similarity determination, if there are multiple standard patterns that are determined to be similar, the standard pattern with the highest degree of similarity is determined as the SL1 similar σ) pattern, and Form a standard pattern -f'. Here, the feature quantities used to calculate the pattern similarity are the maximum image density 1) may, the difference between the red density and the green density (DibDG), and the difference between the green density and the blue density (DG-D).
B), average density D of several parts of the image! , average density of the entire screen 1], histogram of density or hue, spatial frequency distribution, shape of N-valued image, etc. can be considered. Each of these characteristic amounts can be obtained by a known technique, and a pattern of a film image (frame) is constructed by each of these characteristic amounts. Furthermore, the degree of similarity is calculated using the following formula. X=1 to +2 (lΔttGl+IΔG)31)+
3 (lΔULl+lΔI(J,l) +41Δ
CPI 1051ΔI) B l...
・・・・・・・・・(1) Here, K1% has a 5 constant number 1ΔRG l = l (LATD', Iq-IATD
'(G+)-(LATJ)(ladder-hATD((3)
) I・・・・・・・・・(2) 1ΔGB l = l
(LA'l"D'(Q-LATD'(13))-
(LATD(G) LATI)(B) ) l...
...-(3) 1ΔULI = l (DL'-D
U') -(DL-DU) 1・・・・・・・・・・・・
・ (4) )Δf(I, l = l (1) RI'-1) LP'/
)-(DHJ-DLF) l・・・・・・・・・・・・
(5) ・・・・・・・・・・・・ (6) ・・・・・・・・・・・・ (7) (This is the average 101 of the 1# degree difference between adjacent measurement points. , which represents the average contrast of the entire screen.) In addition, LATE') (N) is the total area average transmission density of neutral gray, LA, TD (11 is the total area average transmission density of red, LAT, D ((l) The total area average transmission density of green, LATD (B > is the total area average transmission density of blue, D
min is the minimum density of neutral gray, Dmax is the maximum density of neutral gray, DC is the average density at the center of the screen,]) F is the average density at the periphery of the screen, DU is the average density at the top of the screen, DL is the screen 1) LP is the average density of the lower part, J) is the average density of the right part of the screen, and 1) LP is the average density of the left part of the screen. and,
Those characteristic values with a dash "'" are the characteristic values of the previous frame, and those without a dash are the current frame to be determined as to whether or not the scene is the same as the previous frame. In this manner, in the determination formula (1) above, if the identification signal S is equal to or greater than the fixed value α (identification signal 83 in the above example), it is determined that the previous frame and the current frame are the same scene. In other words. Note that the constant value α can be set arbitrarily,
The similarity level of the identification signal can also be set arbitrarily. On the other hand, as an apparatus for realizing the pattern recognition method according to the present invention as described above, the apparatus shown in FIG. 3 can be considered. ] and is compared with the standard pattern stored in advance in the standard Bakun recording section 14, and the degree of similarity is calculated according to the above equation. Then, the degree of similarity between both patterns calculated by the similarity calculation unit 11 is inputted to the comparison judgment unit 12, and the identification signal S is compared with a predetermined value α, and the comparison result S is Furthermore, when the identification signal S is larger than the predetermined value α, a correction signal CB is sent to the standard pattern modification 1 [section 13 to perform processing to modify the standard pattern, and the identification signal S is output. C1 If it is smaller than the constant value α, it is dissimilar, so the pattern is stored as a standard pattern in the standard pattern storage unit 14. As described above, according to the pattern r+ knowledge method of the present invention, Since we are not trying to determine the similarity between adjacent frames, even if there is a dissimilar frame in the middle of a group of similar frames, the judgment will not be able to remove it and collect similar frames. In addition, we record all patterns once and do not collect mutually fA patterns, and if it is determined that they are similar, we modify and store the standard pattern, so it is possible to memorize it. (It is possible to reduce the number of standard patterns.Furthermore, since the standard patterns are modified, it is possible to improve the similarity with distant frames.In addition, in the case of long films, When a film splice is detected, all standard patterns may be erased and the first pattern of the film may be always stored as a standard pattern.In short, the standard pattern may be stored in advance before film pattern recognition. Just let it happen.

[Brief explanation of drawings]

Figure 1 is a diagram for explaining the conventional pattern recognition method.
FIG. 2 is a diagram for explaining an example of a pattern recognition method according to the present invention, and FIG. 3 is a diagram showing an example of an apparatus for carrying out the present invention. To 7...Standard butter 7.1
0...Pattern detection unit, 11...Similar anti-FtlJQ
Miyako, 12...Comparative publication foot part, J3...Standard pattern correction part, 1.1...Den i, f(/4 pattern memory 4
;.

Claims (1)

[Claims] In a method for sequentially recognizing N consecutively arranged patterns based on similarity, the n (1≦n<N)-th pattern is compared with the (n+1)-th pattern to determine the similarity. If the similarity is less than or equal to the predetermined value, the (n+1)th pattern is stored as the standard pattern, and if the similarity is greater than or equal to the predetermined value, If larger, the nth and (n+1)
Find the standard pattern by averaging the th bar processing turn. A pattern recognition method characterized in that the stored standard patterns are sequentially compared with other patterns, and a new standard pattern is sequentially formed based on the degree of similarity.