Optimization method for inaccuracy of human-scene separation result contour

Abstract

The invention relates to a human-scene separation result contour inaccuracy optimization method comprising the following steps: carrying out edge detection on an input video frame, capturing an area with significant brightness change in an image, and presenting an obvious portrait contour line; after the image subjected to edge detection processing is obtained, filling a blank or fracture area of a portrait contour in the image by using an image restoration technology; setting a connected region screening false detection threshold value, defining the connected region smaller than the false detection threshold value as a misrecognized region, and performing pixel value adjustment on the misrecognized region; sub-region division is carried out on a connected region of the edge region where the portraits and the backgrounds are alternated, pixel values of the sub-regions are checked, brightness and color distribution of the sub-regions are evaluated, and the sub-regions which do not conform to expectation are identified; and correcting the pixel values of the sub-region pixel points. According to the invention, the accuracy of the portrait recognition result can be effectively improved, and the influence of the environment on the recognition result is reduced.

Images