JPS6070321A - Color recognizing apparatus for color picture - Google Patents

Abstract

PURPOSE:To perform effective color judgment, by separating a color picture into (n) pieces of wavelength regions, providing many subdivided small areas for every pattern, selecting the secondary samples based on the sample densities, and computing the modulus for every group. CONSTITUTION:The transmitted light through a picture to be recognized 1 is made to be (r) pieces of electric signals X1-Xr corresponding to wavelengths lambda1-lambdar through an optical system 2, a filter group 3, a photoelectric converter element group 4, and an amplifier group 5. The observed vectors in (r) dimensions are once stored in a memory device 7 through a data processor 6. After the scanning of the entire picture is finished, (m) pieces of the first samples are extracted at random out of the observed vector group in the memory device 7. When the section between the minimum value and the maximum value is divided into l parts, l<r> pieces of subdivided areas are obtained. All samples are included in any of the l<r> pieces of small spaces. The processor 6 sequentially reads all the observed data out of the memory device 7 and imparts the data to a color- recognition operating device 8. Then all the data are divided into a specified number of colors. The results are appropriately combined, and the intended result of color recognition in (n) colors can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a color identification device for a color image, which targets a finite number of colors including brightness and brightness.

Configuration of conventional example and its problems As a color identification device for color images consisting of finite colors, such as maps, dyeing patterns, textile patterns, etc.
A known method that can be realized relatively easily and gives excellent results is a method that captures reflected or transmitted color light as a multispectrum and identifies its pattern. To do this, it is necessary to provide parameters for each image regarding an observation pattern that is statistically reasonable by appropriately determining the identification rule. For this purpose, the process of parameter estimation, ``learning'', is necessary. For learning in a conventional color identification device, usually a color sample whose color is known in advance is read, parameters are calculated, and a classification function is assembled. This kind of "learning" involves preparing a color sample with the colors that appear in the image in addition to the image to be read, or by human intervention to identify areas in the image and their corresponding colors. I need to give instructions. In the former method, factors that greatly control the parameters, such as uneven paint in the target image, are not reproduced in the sample, and in the latter method, the sample area is limited due to human intervention, making it difficult to obtain correct parameters. There were drawbacks.

OBJECTS OF THE INVENTION It is an object of the present invention to eliminate these drawbacks and provide a color identification device for color images equipped with means for easily and objectively obtaining correct parameters. The image is scanned and separated into n wavelength ranges, each of these n patterns is divided into a large number of small spaces, and secondary samples are selected as learning samples according to the sample density of the small spaces. This is a color identification device for color images that calculates a parameter for each group from each group and constructs a classification function.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

Figure 1 shows an example of a target color image, C1, C2...
It is composed of areas of n different colors called C. These colors include uneven coating, uneven printing, and the like.

FIG. 2 is a configuration diagram showing an embodiment of the present invention. In the figure, 1 is an image to be identified consisting of n colors as shown in Fig. 1, and 2 is an optical system for extracting colored light from an observation point, which collects light transmitted through the object surface from a light source (not shown), and or take out the reflected light. The identification target image 1 and the optical system 2 are a group of color separation filters for separating into relative wavelengths using an appropriate scanning system, and 4 is a photoelectric conversion element that converts light components in each wavelength range selected by the color separation filter 3. 6 is an amplifier circuit group, 6 is a theta processor, 7 is a storage device for storing data, and 8 is a color recognition calculation device 4 for determining the color of an observation point.

According to the present invention, colored light from an observation point passes through an optical system 2, a filter group 3, a photoelectric conversion element group 4, and an amplifier group 5,
There are r electrical signals X 1t X2 , . . . xr corresponding to a plurality of wavelength bands (λ1, λ2, .

In other words, the number of colored lights at each observation point is X, x2. ...X
, an r-dimensional observation vector whose components are m pieces (at random or according to predetermined rules)
m >> n , n is the first of the colors to be identified)
Extract the next sample. This is x, x.

73...Represented as 1. Next, Ma'(i-1,2...
-・m), compare each component, and calculate the maximum value of x1max=maxix) I .-=-r) and 1-1nlxZ (k=1, 2...r).

Next, for each component, divide the range between its minimum value and maximum value into one piece.0 As a result, the r-dimensional space in which the sample exists is l
It is divided into r small spaces (equal hypercuboids of each shape). All samples are included in one of these lr small spaces. Among them, i spaces (inn) are selected in descending order of the number of samples they contain. The samples of group i selected in this way are named secondary samples, and each group of secondary samples is regarded as a sample for one different color. The processor 6 examines the components of the samples in each of these i groups and determines the statistical properties of the components. In this way, the discrimination function necessary for the color recognition calculation device 8 is assembled.

The optimal form of the discriminant function can be determined based on the statistical properties of the components. For example, if each component of the sample has a multivariate normal distribution within each group, using a quadratic discriminant function will help Bayesian decision making. The law is fulfilled. In color discrimination, components are not necessarily normally distributed, but for example, a quadratic discriminant function poses no practical problem.

After the discriminant function is resolved in this way, the processor 6 sequentially reads out all observed data from the storage device 7 and supplies it to the color recognition calculation device 8, so that all the data can be rationally combined into one color. Can be divided. i) Since n, the results can be appropriately combined to obtain the desired color recognition results for n colors.

FIG. 3 illustrates the division of the first order sample.

Here, it is assumed that the sample consists of two components xl and x2, and the sample is separated by 7=a. As is clear from the figure, it is easy to find the high-density small spaces 10.11 and 12.13 from the sample data. Assuming that the example in FIG. 3 is an example of three colors, the small space 12 and the small space 13 probably belong to the same color, and it is easy to limit the result to three colors.

In addition, in the previous example, one specimen was mechanically selected in order of density, but as seen in the example in Figure 3, adjacent high-density small spaces often originate from specimens of the same category. . Therefore, as another example, when selecting small spaces in the order of density after dividing into small spaces, the spaces that are adjacent to the already selected small spaces, that is, those whose coordinates are the same except for one, are It is possible to combine them into one.

As another example, when selecting small spaces, it is possible not to select them in order of density, but when focusing on a certain small space, to select a space where the density of all the small spaces adjacent to it is lower than that density. . This is particularly effective when the a priori probability of occurrence of each color varies, or when the target image is one in which the spread of distribution varies greatly for each color.

Effects of the Invention As described above, the present invention scans a color image, separates it into n wavelength ranges, divides each of these n pattern components into a large number of small spaces, and divides the color image into a number of small spaces depending on the sample density of the small spaces. This is a color identification device for color images that selects the next sample as a learning sample, calculates the parameter for each group from each group, and constructs a discriminant function.It has a statistically relatively simple distribution. By performing simple and effective so-called "unsupervised learning" on component patterns, it is possible to realize an effective color identification device for color images.

[Brief explanation of drawings]

FIG. 1 is a plan view showing an example of a color image applied to the present invention, FIG. 142 is a block diagram showing an example of a color identification device according to the present invention, and FIG. FIG. 1...Identification target image, 2...Optical system,
3...Color separation filter group, 4...Photoelectric conversion element group, 6...Amplification circuit group, 6...
Data processor, 7...Storage device, 8...
- Color recognition calculation device, 10-13...Small space.

Claims (1)

[Claims] (1) An optical system that irradiates a color image with light from a light source and separates the local reflected light or transmitted light at an observation point into n wavelength ranges, and a set of these components as a pattern. It is equipped with means for calculating and determining a discriminant function for determining the color of an observation point, and means for scanning a color image, and for each of n pattern components obtained by scanning the color image, between the maximum value and the minimum value. is divided into multiple subintervals, the frequency of appearance of observation points in each subspace in the divided n-dimensional observation space is compared, multiple subspaces in which this value is dominant are selected, and each selected A color identification device for a color image, characterized in that a group of patterns in a small space is used as a learning pattern, a parameter for each group is calculated from each of the groups, and a classification function is configured by the parameter. Among the selected dominant small spaces, two or more small spaces whose ranges are equal except for one are considered to be from the same population, and the population is calculated. A color identification device for a color image according to claim 1. (Hemorrhoids) A patent claim characterized in that the division into small spaces is performed by equally dividing between the maximum value and the minimum value of each pattern component. A color identification device for a color image as described in item 1. 2. The color identification device for a color image according to claim 1, wherein the color images are divided equally.