CN104952049B - Based on the image repair method for facing domain diverging interpolation - Google Patents

Abstract

The invention discloses a kind of image repair method based on field diverging interpolation.The present invention finds eight points that eight directions are nearest around complex point to be repaired first, and further determines that and whether there is empty point in this 8 points, if it is present continuing to spread to outgoing along identical direction, to the last obtains not as eight field points in cavity.Then the single channel value of eight points is subtracted each other two-by-two and seeks absolute value, then the triple channel rgb value of two minimum points of absolute value is averaged respectively, be finally assigned to three passages of complex point to be repaired respectively again, complete image repair.The present invention can preferably repair Fracture Width in the cavity of units pixel coverage, and it is fast to repair speed.

Description

Image Restoration Method Based on Neighborhood Divergence Interpolation

technical field

The invention belongs to the field of graphics and image processing, in particular to an image restoration method based on neighborhood divergence interpolation.

Background technique

With the development of science and technology, 3D technology has become more and more mature. The current mainstream 3D technology mainly converts from 2D to 3D through the reference map and its depth map information, combined with the 3D transformation equation. However, this method has obvious defects, that is, the synthesized new view will have smaller holes and cracks. In order to solve this defect, technicians have studied various hole repair technologies. At present, the main hole repair algorithms include: FMM fast marching algorithm, Criminisi algorithm, neighborhood interpolation algorithm and so on.

The traditional neighborhood interpolation algorithm mainly finds the nearest points in eight directions around the point to be repaired, and then subtracts the single-channel values of the eight points and calculates the absolute value to obtain the two points with the smallest absolute values. Then the three-channel RGB values of these two points are respectively averaged, and the obtained average RGB values are respectively assigned to the points to be repaired, and the image restoration is completed. Experiments show that this method has serious shortcomings in repairing. It can only perfectly repair holes with a crack width less than two pixels, but for holes with a crack width greater than two pixels, the quality of the repaired image is very low.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention proposes a new image restoration method based on neighborhood divergence interpolation.

The technical solution adopted by the present invention to solve its technical problems is as follows:

First, determine the hole point to be repaired, and find the nearest upper left, upper, upper right, left, right, lower left, lower, and lower right points near the hole point to be repaired. They are respectively recorded as: a, b, c, d, e, f, g, h. Then further determine whether there are empty points at the eight points a, b, c, d, e, f, g, and h. If it exists, continue to diverge outward until finding a point that is not a hole. For example: the point a in the upper left direction closest to the hole point to be repaired is a hole point, then continue to traverse the search in the upper left direction until a point that is not a hole is found, which is recorded as a', until finally all eight directions are obtained. There are eight empty points in total, namely a', b', c', d', e', f', g' and h'. Then subtract the single-channel pixel values of the upper left and lower right, upper and lower, upper right and lower left, left and right points for absolute value operation. That is |a'-h'|, |b'-g'|, |c'-f'| and |d'-e'|, and their values are recorded as: diff1, diff2, diff3, diff4. And calculate the minimum value in diff1, diff2, diff3, diff4:

min=minminsize(diff1,minsize(diff2,minsize(diff3,diff4)))

Finally, according to the two points corresponding to the minimum value (for example, if the minimum value is diff1, find points a and h), get the R, G, and B values corresponding to the three-channel map. Namely Ra, Ga, Ba and Rh, Gh, Bh. Then calculate the average value of each channel and three channels at two points, such as R, = (Ra+Rh)/2, G, = (Ga+Gh)/2, B, = (Ba+Bh)/2 . Then assign R, G, and B to the R, G, and B channels of the point to be repaired respectively, and the repair can be completed.

Beneficial effects of the present invention: the present invention is suitable for small cavities with a crack width greater than two pixels, and the repaired image quality is higher than that of traditional methods.

Description of drawings

Fig. 1 is traditional neighborhood interpolation method;

Fig. 2 is the image restoration method of the present invention with divergent interpolation.

detailed description

The present invention will be further described below in conjunction with accompanying drawing:

As shown in Figure 1, the traditional neighborhood interpolation method is to find the points in the eight directions closest to the point to be repaired, regardless of whether there are holes in the eight points (if there are holes, let the single-channel value be zero), The two single-channel values are subtracted to obtain the absolute value, and the two points with the smallest values are obtained, and then the three-channel RGB values are averaged, and assigned to the points to be repaired respectively, and the repair is completed. In this method, if the crack width of the repaired hole is greater than two pixels, there will definitely be a hole point in the eight points around the point to be repaired, resulting in an error in the repair of the RGB three-channel value. However, if we continue to diverge outward to find eight adjacent points that are not holes, and then assign values to the three channels, we can avoid holes and reduce errors, thereby optimizing the image restoration.

As shown in Figure 2, the present invention is mainly based on the neighborhood interpolation algorithm, optimizes and improves, and then improves the image repair quality, and the specific implementation of the present invention is as follows:

Step 1. Determine the hole point to be repaired, and find the nearest upper left, upper, upper right, left, right, lower left, lower, and lower right points near the hole point to be repaired. They are respectively recorded as: a, b, c, d, e, f, g, h.

Step 2. Further determine whether there are empty points at eight points a, b, c, d, e, f, g, and h. If it exists, continue to diverge outward until finding a point that is not a hole. For example: the point a in the upper left direction closest to the hole point to be repaired is a hole point, then continue to search in the upper left direction until a point that is not a hole is found, which is recorded as a'.

Step 3. Repeat step 2, so that the points that are not empty are obtained in the final eight directions, a total of eight points, respectively a', b', c', d', e', f', g' and h'.

Step 4. Subtract the single-channel pixel values of the upper left and lower right, upper and lower, upper right and lower left, left and right points for absolute value operation. That is |a'-h'|, |b'-g'|, |c'-f'| and |d'-e'|, and their values are recorded as: diff1, diff2, diff3, diff4. And calculate the minimum value min=minminsize(diff1,minsize(diff2,minsize(diff3,diff4))) among diff1, diff2, diff3, diff4.

Step 5. Find the two points corresponding to the minimum value (for example, the minimum value is diff1, then find points a and h), and get the R, G, and B values corresponding to the three-channel diagram. Namely Ra, Ga, Ba and Rh, Gh, Bh. Then calculate the average value of each channel and three channels at two points, such as R, = (Ra+Rh)/2, G, = (Ga+Gh)/2, B, = (Ba+Bh)/2 . Then assign R, G, and B to the R, G, and B channels of the point to be repaired respectively, and the repair can be completed.

Claims (1)

1. based on the image repair method for facing domain diverging interpolation, it is characterised in that this method comprises the following steps that：

Step 1. determines empty point to be repaired, and finds cavity point to be repaired nearest upper left, upper, upper right, left and right, left nearby Under, under, the point of bottom right eight；It is designated as respectively：a、b、c、d、e、f、g、h；

Step 2. further determines that eight points of a, b, c, d, e, f, g, h whether there is empty point；If in the presence of continuation outwards diverging Until finding not as the point in cavity；

Step 2 is repeated in step 3. so that all obtains not being empty point on final eight directions, altogether eight points, respectively For a', b', c', d', e', f', g' and h'；

Step 4. is by upper left and bottom right, and upper and lower, upper right and lower-left, the single channel pixel value of the point of left and right, which subtracts each other, seeks absolute value Operation；I.e. | a'-h'|, | b'-g'|, | c'-f'| and | d'-e'|, and its value is designated as respectively：diff1,diff2,diff3, diff4；Calculate diff1, diff2, diff3, the minimum value in diff4；

Two points corresponding to the minimum value that step 5. is looked for, obtain R corresponding to its triple channel figure, G, B values；Then this is calculated respectively The average value of each passage sum of two points, then the average value of three passages of gained is assigned to the threeway of complex point to be repaired respectively Road, you can complete to repair.