JP3088654B2 - Image quality measurement method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image quality measuring method for quantitatively measuring the similarity between an original image and a processed image in a system for digitally processing an image using a computer or the like.

[0002]

2. Description of the Related Art A processed image g for an original image f (x, y)
Various methods have been devised for quantitatively measuring the similarity (so-called image quality) of (x, y). For example, a method of calculating the mean square deviation ∬ (f−g) ² dxdy, a method of calculating the average absolute deviation ∬ | f−g | dxdy, a method of calculating the maximum absolute deviation max | f−g | Have been. In general, these methods are widely used in various fields because their computation methods are relatively simple. Further, the image processing is analyzed from the viewpoint of signal processing, and the processed image g (x, y)
Is assumed to be obtained from a linear operation of the original image f (x, y) and the degradation function h (x, y), and the degradation function h (x, y)
A method for evaluating the image quality of the processed image g (x, y) from the characteristics of the image has been proposed.
(Pages 8 to 171).

[0003]

However, when the image quality is evaluated using the mean square deviation or the mean absolute deviation, a case where a large deviation exists locally and a case where a small deviation exists throughout the entire image. It is difficult to distinguish between Moreover, it is impossible in principle of calculation to eliminate the influence of deviation over the entire image such as illuminance fluctuation. Also,
When the maximum absolute deviation is used, local image fluctuation due to noise or the like greatly affects image quality evaluation. Further, when evaluating the image quality using the deterioration function h (x, y),
What is analyzed is the property of the image processing system itself, that is, the property of the hardware itself.
The image quality of (x, y) is not evaluated.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has as its object the purpose of the present invention as the similarity of a processed image g (x, y) to an original image f (x, y). It is an object of the present invention to provide an image quality measuring method capable of quantitatively evaluating image quality without being affected by illuminance fluctuation, noise, or the like.

[0005]

In order to achieve the above object, according to the present invention, an image input from an ITV camera, that is, an original image, is subjected to image processing, and the image processed image, that is, the processed image is processed. In the image quality measuring method for evaluating the image quality as the similarity with the original image,
A plurality of reference images are prepared, and the plurality of reference images are prepared.
Each of which is based on the quality of any one of the reference images
The image quality as a similarity between each of the plurality of reference images and the processed image is previously calculated quantitatively as a normalized correlation coefficient by a normalized correlation operation , and the number of calculated reference images is calculated. Min normalized correlation coefficient and the above reference
By performing the center of gravity calculation based on the evaluation value of the image, the image quality as the similarity between the original image and the processed image is quantitatively evaluated.

[0006] With this configuration, the evaluation of the image quality of the processed image with respect to the original image is derived based on the normalized correlation coefficient calculated by the normalized correlation operation which is less affected by the illuminance fluctuation. Also, a plurality of images whose image quality has been quantitatively evaluated in advance as an original image are prepared as reference images, and the image quality as the degree of similarity between each of the plurality of reference images and the processed image is calculated. Even when noise is mixed into the original image, an equal amount of noise is mixed into each of the plurality of reference images obtained from the original image, and as a result, each of the plurality of reference images and the processed image The relative evaluation of the image quality as the degree of similarity is such that the influence of noise is eliminated.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. Figure 1 is the original image, Figure 2
(A) to (e) are images obtained when the distance between the original image and the ITV camera is changed in five steps, that is, reference images. Image quality X _k shown in Table 1 (k = a, b, c, d,
e) is the reference image of each of FIGS.
This is a value obtained by subjectively evaluating the image quality (similarity) with respect to the original image shown in FIG. Specifically, since the reference image in FIG. 2C is most similar to the original image shown in FIG. 1, the image quality is set to 100 (X _c = 100) based on this reference image.
FIG. 2A based on the reference image of FIG.
The image quality of each of the reference images (b), (d), and (e) is 30 (X _a = 30), 50 (X _b = 50), and 150, respectively.
( _Xd = 150) and 170 ( _Xe = 170). Here, the closer the image quality is to 100, the more similar to the original image, that is, the higher the image quality is evaluated.

FIG. 3 shows an example of a processed image obtained by subjecting an image input from an ITV camera, that is, an original image, to image processing by an image processing device or the like. A normalized correlation coefficient C _k (k = a, b, c, d, e) between the processed image and the reference images shown in FIGS. 2A to 2E is calculated by a normalized correlation operation shown in the following equation. calculate.

[0009]

(Equation 1)

In equation (1), f _k (i, j) [k =
a, b, c, d, e] are reference images (a) to (a) shown in FIG.
(E) is the density value of each pixel, and g (i, j) is the density value of each pixel of the original image shown in FIG. Where (i,
j) indicates the position of the pixel. The normalized correlation coefficient C _k obtained by the normalized correlation operation shown in the equation (1) is an index for evaluating the degree of similarity between images.
It transitions between +1 and at +1 the image is evaluated to be most similar. Table 1 shows the result of the normalized correlation operation between each reference image shown in FIG. 2 and the processed image shown in FIG.

[0011]

[Table 1]

From Table 1, it can be seen that the reference image shown in FIG. 2 (c) has the highest normalized correlation coefficient with the processed image shown in FIG.

Next, the center of gravity calculation represented by the following equation is performed to calculate the image quality of the processed image shown in FIG.

[0014]

(Equation 2)

The values X _k , which are subjectively evaluated for the image quality of each reference image with respect to the original image, and the normalized correlation coefficient C _k between the processed image and each reference image shown in Table 1, are substituted into the equation (2). Then, the image quality X of the processed image shown in FIG. 3 is evaluated as 91.30, and it is determined that the similarity of the processed image shown in FIG. 3 with the original image is relatively high.

Here, considering the case where noise is mixed in the original image, an equal amount of noise is mixed into each of a plurality of reference images obtained from the original image. FIG. 2 based on the reference image of FIG.
The evaluation of the image quality of each of the reference images (a), (b), (d), and (e) is the same as in the case where no noise is mixed, and as a result, each of the plurality of reference images and the processed image are compared. The evaluation of the image quality as the degree of similarity is not affected by the presence or absence of noise.

In the above embodiment, the image quality X _k of the reference image with respect to the original image is expressed as a subjective evaluation based on an arbitrary one of a plurality of reference images. May be expressed as a parameter between customers as a parameter.

[0018]

As described above, according to the present invention, the original image
Prepare a plurality of reference images as
Each of the reference images is based on the quality of any one of the reference images.
The image quality as the degree of similarity between each of the plurality of reference images and the processed image obtained by performing image processing on the original image is normalized as a normalized correlation coefficient. Each is calculated by calculation, and the calculated reference
Of normalized correlation coefficients for the number of reference images and evaluation of the reference image
By calculating the center of gravity based on the values, the image quality as the similarity between the original image and the processed image is quantitatively evaluated. It is derived based on the normalized correlation coefficient calculated by the normalized correlation operation which is less likely to be received. This makes it possible to eliminate the influence of deviation over the entire image such as illuminance fluctuation.For example, it is also possible to separately evaluate a case where a large deviation exists locally and a case where a small deviation exists throughout the image. became.

Further, even when noise is mixed in the original image,
The same amount of noise will be mixed into each of the plurality of reference images obtained from the original image, and as a result, the evaluation of the image quality as the similarity between each of the plurality of reference images and the processed image will be described. Since the influence of noise has been eliminated, it is no longer necessary to remove noise in preprocessing.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example of an original image used for explaining an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a reference image when the original image of FIG. 1 is photographed at a different distance from an ITV camera.

FIG. 3 is a diagram illustrating an example of a processed image that has been subjected to image processing.

Claims (1)

(57) [Claims] An image quality measuring method for performing image processing on an image input from an ITV camera, ie, an original image, and evaluating the image quality as a degree of similarity between the image-processed image, ie, the processed image and the original image. A plurality of reference images are prepared as the original images, and each of the plurality of reference images is replaced with an arbitrary one of the reference images.
The advance quantitatively evaluate the image quality as a reference, the image quality of a similarity with each of the processed image of the plurality of reference images calculated by the normalized correlation operation as a normalized correlation coefficient is calculated Normalized correlation coefficients for the number of reference images
By calculating the center of gravity based on the evaluation value of the illumination image ,
An image quality measuring method characterized by quantitatively evaluating the image quality as a similarity between an original image and a processed image.