JPH06295168A - Method for detecting color unevenness on display element - Google Patents

Abstract

PURPOSE:To stably detect only the color unevenness part on a display element in a picture by using a color information processing system close to a human visual sense characteristic. CONSTITUTION:An inspected display element 1 is image picked up by a color camera 2, and red R, green G, blue B of three primary colors are converted into a luminance signal R+G+B=L and chrominance signals G-R=P, B-(R +G)=Q by a chrominance conversion circuit 3. After the L, P, G are converted to the digital data, are divided to the area with a uniform value related into L, P, G images by histogram process/area division processors 10, 11, 12 through image memories 7, 8, 9 to be stored in an integration division result memories 13, 14, 15. The division area is integrated by an area integration processor 16 to be stored in an area integration result 17. The divided areas are integrated by an area integration processor 16 to be stored in an area integration memory 17. A neural network making the integrated areas (uniform areas of L, P, Q) nodes is formed, and the color unevenness part is detected by an extraction processor 18.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting color unevenness in a color cathode ray tube, a liquid crystal display device or the like.

[0002]

2. Description of the Related Art As a method for detecting color unevenness of a liquid crystal display element, there is a "uniformity state inspection method" described in JP-A-2-1932271. This method is a method in which the color unevenness portion is emphasized and detected by calculating the difference between the image obtained by shifting the display element vertically, horizontally, and diagonally with respect to the image captured by a camera or the like.

[0003]

In the above-mentioned conventional technique, the color unevenness portion is emphasized and detected by calculating the difference between the image obtained by shifting the display element vertically, horizontally, and diagonally with respect to the image taken by the camera or the like. Therefore, it is difficult to realize stable color unevenness extraction because the degree of emphasis of the color unevenness portion changes depending on the shift amount and the direction. In addition, there is a problem in that the uneven color portion cannot be emphasized when another uneven color portion exists in the shifted portion. An object of the present invention is to provide a method for detecting color unevenness of a display device such as a color cathode ray tube or a liquid crystal, and further to provide a method for detecting color unevenness which is close to and stable to human visual characteristics.

[0004]

In order to achieve the above object, according to the present invention, a display device for displaying an image on a display screen by a color display element, the display screen is imaged by a color sensor, Three outputs of three primary colors, red (R), green (G), and blue (B), which are obtained by the color sensor, are output as one light / dark signal R + G + B and two color signals G-R, B- (R).
+ G), the three signals are converted into three types of digital data, the three types of digital data are stored in image memories, respectively, and the color unevenness of the display element is detected using the three types of digital data. In doing so, means for dividing each of the image memories into areas in which the value of the data stored in the image memory is within a predetermined range, and the values of the three types of data are respectively determined based on these areas. A hop field type in which an area on the image memory divided into areas within the range is considered as one node, and the node output increases as the node deviates from the reference color that should be the original color. Means for constructing a neural network, means for updating the network state by iterative calculation relating to the deviation from the reference color, and the network reaching a stable equilibrium state. Once the by extracting the output is large nodes may comprise means for detecting only the color unevenness portion.

[0005]

Function: An image of a display element to be inspected is picked up by a color sensor such as a color camera, and the three primary color outputs of the color sensor, that is, red (R),
By converting green (G) and blue (B) into a light / dark signal R + G + B and two color signals G-R and B- (R + G), they have no relation to colors and play the most important role in human vision. It is possible to obtain two color signals G-R (difference between green and red) and B- (R + G) (difference between blue and yellow), which are in a complementary color relationship that has a high human color discrimination ability. Therefore, it is possible to construct a color information processing system that is close to human visual characteristics. The human visual characteristics are described in detail in "Eyes and Sensors of Living Things" (edited by Yoshishige Shimizu, Information Research Society).

Further, in the picked-up image of the display element which is turned on in a certain standard color, the R + G + B, GR, B-
Of the areas where the value of (R + G) is uniform, only the area where the deviation from the reference color (the original color) is large can be extracted in the stable equilibrium state of the Hopfield neural network, so that only the uneven color areas can be stably extracted. Can be detected.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS.

FIG. 1 is a diagram showing a system configuration of a color unevenness detecting device for a display element for realizing the present invention. The display element 1 to be inspected is imaged by the color camera 2 in a state where one surface of the CRT is uniformly displayed in white or in a reference color (reference color). The three primary color outputs of the color camera 2, that is, red (R), green (G), and blue (B) are output by the color conversion circuit 3 composed of an adder / subtractor using an operational amplifier, and a bright / dark signal R + G + B and two signals. Color signals GR, B-
Converted to (R + G). Hereinafter, R + G + B, GR,
B- (R + G) will be called L, P, and Q.

L, P and Q are converted into digital data by A / D converters 4, 5 and 6, and image memories 7, 8 and 9 are obtained.
Stored in. The digital data after A / D conversion is L,
A configuration in which P and Q are converted and stored in the image memory is also possible. As the color conversion circuit 3 in this case, a Progr-ammable Color Space Converter Bt281 manufactured by Brooktree, USA can be used.

The data in the image memories 7, 8 and 9 are divided by the histogram processing / region division processors 10, 11 and 12 into regions having uniform values for each of the L, P and Q images, and the result is the region division result. Memories 13, 14, 15
Stored in. By the area integration processor 16, L,
The P and Q divided areas are integrated and the result is stored in the area integration result memory 17. The integrated areas are L, P, and Q, respectively.
Is a uniform area, and in order to extract uneven color areas,
The uneven color extraction processor 18 is provided.

FIG. 2 is a flowchart showing the processing procedure of the color unevenness detection processing of the present invention. The outline of the processing procedure is shown below.

The display image to be inspected taken in by a color sensor such as a color camera is converted into L, P and Q images by the color conversion circuit 3 and stored in the image memories 7, 8 and 9 (step S1).

For the L, P, and Q images, the histogram processing / region dividing processors 10, 11, and 12 generate histograms (frequency distribution of image values) (step S2).

Histogram processing / region division processor 1
The histogram is evenly divided by 0, 11, and 12 (step S3).

Based on step S3, the histogram processing / region division processors 10, 11, 12 cause L,
The P and Q images are individually divided into regions. It can be considered that the divided regions (segments) have uniform L values in the L image, uniform P values in the P image, and uniform Q values in the Q image (step S4).

By the area integration processor 16, L, P,
Integration processing of the divided segments of the Q image is performed. The integration processing is to make the pixels a and b belong to the same segment when the following conditions 1 to 3 are satisfied for the pixels a and b.

Condition 1: L segment to which pixel a belongs = L segment to which pixel b belongs Condition 2: P segment to which pixel a belongs = P segment to which pixel b belongs Condition 3: Q segment to which pixel a belongs = Q to which pixel b belongs It can be considered that the values of L, P, and Q of each obtained segment are uniform (step S5).

With respect to the segment obtained in step S5, the color unevenness extraction processor 18 calculates an equivalent ellipse using the principal component analysis (step S6).

The color shading extraction processor 18 makes the segment a node, and the input of the node is proportional to the difference between the value at the L, P, and Q nodes and the value at the reference color (the original color), The integration between the two forms a Hopfield neural network defined by the proximity calculated from the equivalent ellipse in step S6 and the difference between the nodes of the L, P, and Q values (step S7). The Hopfield neural network is described in detail in "Neural Network Information Processing" (Hideki Aso, Sangyo Tosho).

The color shading extraction processor 18 repeats the network calculation from an appropriate initial value, and at the stage of convergence, only the node having a large output is extracted. The segment corresponding to the extracted node corresponds to the uneven color portion (step S8).

The details of each process will be described below.

Area extraction by histogram uniform division will be described with reference to FIG. As shown in the figure, the histogram is divided at equal intervals Δ so as to end at the valley that crosses the peaks of the histogram. Since it ends in a valley, it is not strictly equal, but becomes equal in the vicinity of a mountain, that is, the value of an image with a high appearance frequency. FIG. 3 shows an example in which a region (hatched portion) having a value of an image near the maximum mountain in the histogram is extracted.

FIG. 4 shows an example in which the L, P, and Q images are divided into regions based on the histogram equal division. In this example,
The L, P, and Q images are divided into 7, 5, and 5 segments, respectively. Further, in the figure, 39 segments are generated by performing the above-mentioned segment integration processing. The obtained 39 segments become a uniform region in which the values of L, P, and Q have a variation of about Δ described above.

FIG. 5 is a diagram for explaining the equivalent elliptic approximation of a segment. The coordinate value of the pixel belonging to the segment is (x
_1, y _1), ......, and (x _0, y _0), obtaining the variance-covariance matrix B regarding the coordinates of (Equation 1).

[0025]

[Equation 1]

The direction of the eigenvector of the matrix B is the axial direction of the equivalent ellipse, and the square root of the eigenvalue is the axial length. This method is called principal component analysis, and the axis of the equivalent ellipse corresponds to the principal component axis.
For more information on principal component analysis, please refer to "Sequential multivariate analysis" (Tadakazu Okuno et al., Japan Science and Technology Federation).

The proximity P _{x of the} segment is defined on the straight line connecting the centers of the equivalent ellipses as shown in FIG. The ratio of the distance | O ₁ O ₂ | between the intersections of the straight line connecting the centers of the equivalent ellipses and the ellipse to the distance between the centers of the equivalent ellipses | P ₁ P ₂ | The closer the segments are, the smaller the proximity is.

FIG. 7 is a diagram showing an example of the construction of the color spot extraction Hopfield neural network having 39 segments as nodes obtained in FIG. The input of the node is a linear function relating to the difference between the L, P, Q values of the reference color (the original color) and the average value of the L, P, Q of the node as in (Equation 2).

[0029]

[Equation 2]

The weighting factor to be applied to the connection between nodes is calculated by the above _equation (Equation 3) when there is no connection when the proximity P _x is larger than the proximity threshold P _{th and} when it is P _th or less.

[0031]

[Equation 3]

The updating rule of the output of the node is performed in accordance with (Equation 5) in which the first-order differential equation of (Equation 4) is discretized.

[0033]

[Equation 4]

[0034]

[Equation 5]

FIG. 8 is a diagram showing a segment corresponding to a node having a large output after iterative calculation according to (Equation 5) in the initial state of the network constructed under the following conditions. Figure 8
The segment of corresponds to a portion having a large deviation from the reference color, that is, an uneven color portion.

K '/ K = 0.5 L, P, Q thresholds are 7%, 3%, 3% of full scale.
% Proximity threshold P _th = 50% τ = 10 Initial output of node = 0 Initial curtain potential of node = 0 As described above, according to the present embodiment, a display device such as a color cathode ray tube or a liquid crystal is targeted. Since a color information processing system that is close to human visual characteristics can be used and only the uneven color portion can be emphasized and extracted, there is an effect that the uneven color detection can be performed stably.

[0037]

According to the present invention, it is possible to construct a color information processing system close to human visual characteristics in detecting color unevenness of a display device such as a color CRT or liquid crystal.
There is an effect that human sensory evaluation can be quantified. Also,
Since the uneven color portion can be emphasized by the Hopfield type neural network processing, there is an effect that even a slight uneven color can be stably detected.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of an apparatus for detecting color unevenness of a display element, which is an embodiment of the present invention.

FIG. 2 is a diagram showing a flow of color unevenness detection processing of the present invention.

FIG. 3 is a diagram illustrating region extraction by histogram uniform division.

FIG. 4 is a diagram illustrating region division and segment integration processing.

FIG. 5 is a diagram illustrating an equivalent elliptic approximation of a segment.

FIG. 6 is a diagram that defines the proximity of segments.

FIG. 7 is a diagram showing a configuration example of a color shading extraction Hopfield network having segments as nodes.

FIG. 8 is a diagram showing an extracted uneven color portion.

[Simple explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Inspected display element, 2 ... Color camera, 3 ... Color conversion circuit, 4-6 ... A / D converter, 7-9 ... Image memory, 10
... 12 ... Histogram processing / region division processor, 13
-15 ... Area division result memory, 16 ... Area integration processor, 17 ... Area integration result memory, 18 ... Color spot extraction processor.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Internal reference number FI Technical display location H04N 9/12 A 9187-5C 17/04 C 6942-5C

Claims (1)

[Claims] 1. A display device for displaying an image on a display screen by means of a color display element, wherein the display screen is imaged by a color sensor, and three primary color outputs, red (R) and green (, are obtained by the color sensor. G) and three outputs of blue (B), one light / dark signal R + G + B and two color signals GR, B- (R +
G), the three signals are converted into three types of digital data, the three types of digital data are stored in image memories, respectively, and the color unevenness of the display element is detected using the three types of digital data. In doing so, each of the image memories is divided into areas in which the value of the data stored in the image memory is within a predetermined range, and based on these areas, the values of the three types of data are each within the predetermined range. Hopfield neural network that considers a region on the image memory divided into certain regions, defines the region as one node, and has a larger node output as the node deviates from the reference color that should be the original color. And the network state is updated by iterative calculation related to the deviation from the reference color, and when the network reaches a stable equilibrium state, the output is large. A method for detecting color unevenness in a display element, which comprises detecting only a color unevenness portion by extracting a color code.