| CPC G06T 7/0012 (2013.01) [G06T 2207/30088 (2013.01)] | 5 Claims |
|
1. A machine vision-based method for rapidly detecting a scar contour,
comprising the following steps:
collecting to-be-detected neck images of a number of persons, wherein the to-be-detected neck images contain a plurality of superpixel blocks, each superpixel block contains a plurality of pixel points, and each pixel point corresponds to an L-channel value, an a-channel value, and a b-channel value under a Lab space;
obtaining a scar color difference salient degree and a scar brightness difference salient degree of each superpixel block according to a distribution difference situation of L-channel values and a-channel values among different superpixel blocks, wherein the scar color difference salient degree is used to describe a difference between a scar color in the superpixel block and an overall color in other superpixel blocks, and the scar brightness difference salient degree is used to describe a difference between a scar brightness in the superpixel block and an overall brightness in the other superpixel blocks;
obtaining a contour similarity degree of each superpixel block according to a correlation influence situation between the scar brightness difference salient degree and the scar color difference salient degree of the superpixel block, and the b-channel value of each pixel point;
and performing scar contour detection according to the contour similarity degree;
wherein a specific method comprised in obtaining the contour similarity degree of each superpixel block according to the correlation influence situation between the scar brightness difference salient degree and the scar color difference salient degree of the superpixel block, and the b-channel value of each pixel point is as follows:
obtaining a scar color difference salient weight and a scar brightness difference salient weight of an i2-th superpixel block and an i3-th superpixel block according to a difference situation of scar brightness difference salient degrees and scar color difference salient degrees between the i2-th superpixel block and the i3-th superpixel block:
obtaining a regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block according to the scar color difference salient weight and the scar brightness difference salient weight of the i2-th superpixel block and the i3-th superpixel block, and the b-channel value of each pixel point;
obtaining a contour similarity factor of the i2-th superpixel block and the i3-th superpixel block according to the regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block; and
acquiring a contour similarity factor of all any two adjacent superpixel blocks, linearly normalizing all contour similarity factors, and recording each normalized contour similarity factor as the contour similarity degree;
wherein a specific method comprised in obtaining the scar color difference salient weight and the scar brightness difference salient weight of the i2-th superpixel block and the i3-th superpixel block according to the difference situation of the scar brightness difference salient degrees and the scar color difference salient degrees between the i2-th superpixel block and the i3-th superpixel block is as follows:
recording an absolute value of a difference of the scar color difference salient degrees between the i2-th superpixel block and the i3-th superpixel block as a color difference salient difference value of the i2-th superpixel block and the i3-th superpixel block; recording a sum of 1 and the color difference salient difference value as the scar color difference salient weight of the i2-th superpixel block and the i3-th superpixel block;
recording an absolute value of a difference of the scar brightness difference salient degrees between the i2-th superpixel block and the i3-th superpixel block as a brightness difference salient difference value of the i2-th superpixel block and the i3-th superpixel block; and
recording a sum of 1 and the brightness difference salient difference value as the scar brightness difference salient weight of the i2-th superpixel block and the i3-th superpixel block;
wherein a specific method comprised in obtaining the regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block according to the scar color difference salient weight and the scar brightness difference salient weight of the i2-th superpixel block and the i3-th superpixel block, and the b-channel value of each pixel point is as follows:
recording a square of a difference between a mean of L-channel values of all pixel points in the i2-th superpixel block and a mean of L-channel values of all pixel points in the i3-th superpixel block as a third difference of the i2-th superpixel block and the i3-th superpixel block: recording a product of the third difference and the scar brightness difference salient weight of the i2-th superpixel block and the i3-th superpixel block as a third product of the i2-th superpixel block and the i3-th superpixel block;
recording a square of a difference between a mean of a-channel values of all the pixel points in the i2-th superpixel block and a mean of a-channel values of all the pixel points in the i3-th superpixel block as a fourth difference of the i2-th superpixel block and the i3-th superpixel block: recording a product of the fourth difference and the scar color difference salient weight of the i2-th superpixel block and the i3-th superpixel block as a fourth product of the i2-th superpixel block and the i3-th superpixel block:
recording a square of a difference between a mean of b-channel values of all the pixel points in the i2-th superpixel block and a mean of b-channel values of all the pixel points in the i3-th superpixel block as a fifth difference of the i2-th superpixel block and the i3-th superpixel block;
recording a sum of the third product, the fourth product and the fifth difference as a first sum value of the i2-th superpixel block and the i3-th superpixel block; and recording an arithmetic square root of the first sum value as the regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block:
wherein a specific method comprised in obtaining the contour similarity factor of the i2-th superpixel block and the i3-th superpixel block according to the regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block is as follows:
recording an absolute value of a difference of the scar brightness difference salient weight and the scar color difference salient weight between the i2-th superpixel block and the i3-th superpixel block as a first absolute value of the i2-th superpixel block and the i3-th superpixel block: recording a sum of 1 and the first absolute value as an adjustment factor of the i2-th superpixel block and the i3-th superpixel block: recording a product of the adjustment factor and a Euclidean distance of centers of gravity between the i2-th superpixel block and the i3-th superpixel block as a fifth product of the i2-th superpixel block and the i3-th superpixel block; and recording a ratio of the fifth product to the regional chromaticity distance between the i2-th superpixel block and the i3-th superpixel block as the contour similarity factor of the i2-th superpixel block and the i3-th superpixel block;
wherein a specific method comprised in performing the scar contour detection according to the contour similarity degree is as follows:
presetting a contour similarity degree threshold T1, merging, if a contour similarity degree of an i2-th superpixel block and an i3-th superpixel block is less than T1, the i2-th superpixel block with the i3-th superpixel block until a contour similarity degree of all any two adjacent superpixel blocks is greater than or equal to T1, and acquiring all the superpixel blocks in the to-be-detected neck image: inputting all the superpixel blocks into a CA saliency detection algorithm to acquire a saliency map; and
presetting a salient value threshold T2, recording a pixel point with a salient value greater than T2 in the saliency map as a scar pixel point, acquiring all scar pixel points, and recording an image region occupied by all the scar pixel points as a scar contour region.
|