Research On Interactive Matting Method For High Resolution And Large Size Images

Digital image matting is widely used in many fields,especially in the field of film and television production and virtual reality.So far,there are many traditional matting algorithms and deep learning matting algorithms.Due to the huge demand for computation and memory,it is difficult to process high-resolution and large-size images on popular hardware(desktop computers,portable computers,mobile phones and other devices),whether it is traditional algorithm matting or deep learning algorithm matting.This thesis does not consider the hardware solution to this problem,but proposes an interactive matting method for high-resolution large-scale images from the perspective of software,which can be applied to ordinary devices.Its core idea is to change the big picture into the small one,to change the big picture into the small one and to act serially.The method consists of two core parts,namely,the segmentation algorithm guided by the target contour and the distance weighted alpha image splicing algorithm.Large image segmentation algorithm guided by target contour mainly aims at the image with simple target contour.It can automatically generate guidance map by recording and processing the contour path of target area described by users,and segment the unknown category areas in guidance map and original large image into small blocks along the target contour.The small guide image and the corresponding original image can be sent into any existing matting algorithm,and matting can be realized by ordinary hardware.Distance weighted alpha image splicing algorithm is used to realize the splicing of several small matting results(alpha image)into the original image size.Because there are overlapping areas between segmented small images and between small matting results(to compensate for the matting accuracy degradation of small matting results near the boundary),this thesis proposes a distance based splicing weight calculation method,and combines multiple overlapping matting results at any position into one final result based on the distance weight.The method is verified on a common computer,and the results show that it can easily handle high-resolution large-scale image matting.Experimental results on standard matting datasets show that compared with the baseline method of global matting,the proposed method does not bring significant performance degradation...