CN104952049A

CN104952049A - Image inpainting method based on domain divergence interpolation

Info

Publication number: CN104952049A
Application number: CN201510348382.1A
Authority: CN
Inventors: 杨柏林; 熊渝
Original assignee: Zhejiang Gongshang University
Current assignee: Zhejiang Gongshang University
Priority date: 2015-06-23
Filing date: 2015-06-23
Publication date: 2015-09-30
Anticipated expiration: 2035-06-23
Also published as: CN104952049B

Abstract

The invention discloses an image inpainting method based on domain divergence interpolation. The method comprises the following steps: eight nearest points in eight directions around a to-be-inpainted point are found out firstly, whether a hole point exists in the eight points is further determined, and if the hole point exists, divergence is performed continuously in the same direction until eight non-hole domain points are obtained finally; single channel values of the eight points are subjected to subtraction pairwise, absolute values are solved, three-channel RGB (red, green, blue) values of two points with minimum absolute values are averaged respectively, finally, values are assigned to three channels of the to-be-inpainted point respectively, and image inpainting is finished. The image inpainting method can better repair holes with the crack width in the single-digit pixel range, and the repair speed is high.

Description

The image repair method of interpolation is dispersed based on field

Technical field

The invention belongs to graph and image processing field, particularly a kind of image repair method dispersing interpolation based on field.

Background technology

Along with the development of science and technology, 3D technology is more and more ripe.The 3D technology of current main flow, mainly by reference to figure and depth map information thereof, in conjunction with 3D transformation equation, forwards 3D to from 2D.But this method, can exist obvious defect, can there is less empty crack in the new view namely synthesized.In order to solve this defect, technician have studied various empty recovery technique, and empty restore design main at present comprises: FMM fast marching algorithms, Criminisi algorithm, field interpolation algorithm etc.

Traditional field interpolation algorithm, mainly by finding the point that around complex point to be repaired, eight directions are nearest, the single channel value of these eight points is made to subtract each other between two and ask absolute value again, obtain two points that absolute value is minimum, again these two some triple channel rgb values are averaged respectively, the average RGB value obtained is distinguished assignment to complex point to be repaired, namely completes image repair.Experiment shows that the method reparation exists serious deficiency, the cavity that can only perfect repair Fracture Width and be less than two pixels, and Fracture Width is greater than to the cavity of two pixels, its mass effect of repairing image very low.

Summary of the invention

For the deficiencies in the prior art, the present invention proposes a kind of image repair method dispersing interpolation based on field newly.

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

First, determine empty point to be repaired, and find nearest upper left near the point of cavity to be repaired, upper, upper right, left and right, lower-left, under, eight, bottom right point.Be designated as respectively: a, b, c, d, e, f, g, h.And then determine whether a, b, c, d, e, f, g, h eight points exist empty point further.If exist, then continue outwards to disperse until finding is not the point in cavity.Such as: the some a of the upper left nearest apart from cavity to be repaired point is empty point, then continue to find to upper left traversal, until searching out not is the point in cavity, is designated as a, until final eight directions all obtain not be the point in cavity, amounts to eight points.Then by upper left and bottom right, upper and lower, upper right and lower-left, the single channel pixel value of the point of left and right subtracts each other asks absolute value operation.Namely | a-h|, | b-g|, | c-f|, | d-e|, and its value is designated as respectively: diff1, diff2, diff3, diff4.And calculate diff1, minimum value in diff2, diff3, diff4:

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

Finally according to two points corresponding to minimum value, (such as minimum value is diff1, then find an a and h), obtain the R that its triple channel figure is corresponding, G, B value.I.e. Ra, Ga, Ba and Rh, Gh, Bh.Then calculate respectively two 3 each passages of passage and mean value, as R ,=(Ra+Rh)/2, G ,=(Ga+Gh)/2, B ,=(Ba+Bh)/2.Again by R, G, B, assignment gives the R of complex point to be repaired respectively, and G, B triple channel, can complete reparation.

Beneficial effect of the present invention: the present invention is a kind of little cavity being applicable to Fracture Width and being greater than more than two pixels, compares its picture quality of repairing of classic method more high.

Accompanying drawing explanation

Fig. 1 is traditional field interpolation method;

Fig. 2 is that the field of the invention disperses interpolation image restorative procedure.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described:

As shown in Figure 1, traditional field interpolation method be find complex point to be repaired nearest eight directions on point, no matter whether there is cavity in these eight points (if there is cavity, its single channel value is then made to be zero), all single channel value is between two subtracted each other and ask absolute value, obtain two points that value is minimum, then its triple channel rgb value is averaged, assignment gives complex point to be repaired respectively, namely completes reparation.The method, if the Fracture Width repairing cavity is greater than two pixels, can make definitely can there is empty point in around complex point to be repaired eight points, thus causes RGB triple channel to be worth repairing error.And if know that finding is not eight field points in cavity when looking for eight points in field by constantly outwards dispersing, then carry out triple channel and be worth assignment, can avoid occurring cavity, reduce error, thus optimized reparation image.

As shown in Figure 2, the present invention, mainly according to field interpolation algorithm, optimizes improvement, and then improves image repair quality, and specific embodiment of the invention is as follows:

Step 1. determines empty point to be repaired, and find nearest upper left near the point of cavity to be repaired, upper, upper right, left and right, lower-left, under, eight, bottom right point.Be designated as respectively: a, b, c, d, e, f, g, h.

Step 2. determines whether a, b, c, d, e, f, g, h eight points exist empty point further.If exist, then continue outwards to disperse until finding is not the point in cavity.Such as: the some a of the upper left nearest apart from cavity to be repaired point is empty point, then continue to find to upper left traversal, until searching out not is the point in cavity, is designated as a.

Step 3. carry out step 2 repeatedly, and making final eight directions all obtain not is the point in cavity, amounts to eight points.

Step 4. is by upper left and bottom right, upper and lower, upper right and lower-left, and the single channel pixel value of the point of left and right subtracts each other asks absolute value operation.Namely | a-h|, | b-g|, | c-f|, | d-e|, and its value is designated as respectively: diff1, diff2, diff3, diff4.And calculate diff1, minimum value min=minminsize (diff1, minsize (diff2, minsize (diff3, diff4))) in diff2, diff3, diff4.

(such as minimum value is diff1 to two points corresponding to step 5. minimum value of looking for, then find an a and h), obtain the R that its triple channel figure is corresponding, G, B value.I.e. Ra, Ga, Ba and Rh, Gh, Bh.Then calculate respectively two 3 each passages of passage and mean value, as R ,=(Ra+Rh)/2, G ,=(Ga+Gh)/2, B ,=(Ba+Bh)/2.Again by R, G, B, assignment gives the R of complex point to be repaired respectively, and G, B triple channel, can complete reparation.

Claims

1. disperse the image repair method of interpolation based on field, it is characterized in that the concrete steps of the method are as follows:

Step 1. determines empty point to be repaired, and find nearest upper left near the point of cavity to be repaired, upper, upper right, left and right, lower-left, under, eight, bottom right point; Be designated as respectively: a, b, c, d, e, f, g, h;

Step 2. determines whether a, b, c, d, e, f, g, h eight points exist empty point further; If exist, then continue outwards to disperse until finding is not the point in cavity;

Step 3. carry out step 2 repeatedly, and making final eight directions all obtain not is the point in cavity, amounts to eight points;

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

Two points that step 5. minimum value of looking for is corresponding, obtain the R that its triple channel figure is corresponding, G, B value; Then calculate respectively these two each passages of point and mean value, then by the mean value of three of gained passages respectively assignment give the triple channel of complex point to be repaired, can reparation be completed.