Research On Optimization Algorithm Of Image Color Layering

In digital image editing,the image is usually composed of layers,which are used to separate and represent different parts of the image.Layers facilitate the operation of the image,for example,the user can add or delete layers,can also change the hue of the layer,layer replacement and other operations,to facilitate the realization of image editing.However,once the image is rasterized,the layer no longer exists.In this paper,a layer is defined as an image segment with approximately uniform color.The purpose of this paper is to decompose a given image into several high-quality color layers to facilitate image editing.The main work of this paper is as follows:(1)Aiming at the hierarchical optimization objective function,four sparse regularization methods are compared and analyzed.Layer decomposition is a typical ill posed problem,and its solution is generally ill posed.In fact,in high-quality layer decomposition results,the alpha value of each pixel corresponding to each layer is often sparse.Based on this prior knowledge,sparse regularization is an effective method to narrow the range of feasible solutions or improve the stability of solutions.In this paper,four sparse regularization terms are added to the objective function of basic color layer decomposition,which can constrain the solution of the objective function and make its value tend to 0 or 1,so as to improve the effect of layer decomposition.These four kinds of sparse regularization are:L1 regularization,L2 regularization,Levin regularization and YAGIZ regularization,and the corresponding solutions are given in the end of this paper.In this paper,the layer decomposition method with sparse regularization is verified by CIR data set.The experimental results show that the alpha value of the objective function with Levin regularization is the best,followed by L1 regularization,L2 regularization is relatively poor,and YAGIZ regularization is the worst;In the interior and boundary of the layer,Levin regularization has the best color homogeneity,and the boundary is too natural smooth.L1 regularization takes the second place,L2 regularization is relatively poor,and YAGIZ regularization is the worst.(2)A layered method based on deep network is implemented.In order to solve the efficiency problem of traditional methods,a layer decomposition method based on U-Net and SPP net is proposed.Based on the existing layer decomposition method based on U-Net,a spatial pyramid pooling module is added to effectively aggregate the context information of different feature map regions.The encoded information is pooled by spatial pyramid and integrated into the decoder.Experimental results show that the layer decomposition method based on U-Net and SPP net proposed in this paper obtains high quality layer decomposition results,and is superior to the layer decomposition method based on U-Net in image reconstruction measurement,and also improves the operation speed compared with the traditional method.