Research On Image Segmentation Method Based On Spectral Hashing

Image segmentation is one of the primary research in computer vision and plays a vital role in image processing.With the updating of digital products,digital images have appeared more and more in our lives.For image denoising,image scaling,image classification,and image understanding applications,accurate image segmentation as basic research play a vital role.With the emergence of more and more kinds of images,new challenges and requirements are put forward for image segmentation algorithms.In this paper,we propose an image segmentation algorithm based on multi-view feature semantics for complex structure images inspired by human perception research.We combine the semantic hash model to accelerate the optimization process and achieve ideal segmentation results.Complex structure image is a kind of image with "intra-class difference and inter-class similarity".In the process of such image segmentation,the traditional segmentation methods will have some problems,such as low accuracy,over-segmentation,under-segmentation,difficult manual adjustment and so on.In this paper,after analyzing the characteristics of complex structure images,a similarity measure function based on the semantics of pixel-level features is proposed.By comparing the similarity distance between two pixels,the similarity matrix of images is obtained,and the image segmentation algorithm based on the clustering idea is realized.To solve the problem of sizeable computational overhead and long time-consuming,a semantic hash model is introduced to accelerate the optimization process,and the final segmentation result is obtained by calculating the Hamming distance.Based on the research results of human perception,different view features(saliency,depth of field,color,etc.)contribute differently to different types of images.Traditional segmentation algorithm combines multi-view features into long vectors,without considering the separate contribution of various view features to different images.To solve this problem,this paper proposes a multi-view feature segmentation algorithm.In image multi-view feature space,by minimizing feature similarity distance function,feature selection matrix is proposed to find the optimal feature projection plane,so that multi-view feature projection has the best image segmentation effect.Compared with the latest unsupervised image segmentation algorithm and a deep learning algorithm based segmentation algorithm,the effectiveness and advancement of the proposed algorithm are verified.The main innovations of this paper are as follows:(1)A pixel-level similarity measure function is proposed for complex structure images,and the image segmentation model is represented as the problem of minimizing the similarity of pixel pairs.In the optimization process,the semantic hash model is introduced to accelerate the optimization process.(2)Inspired by the research of human perception,a multi-view feature semantic image segmentation algorithm is proposed.According to the multi-dimensional features of the image,the optimal projection plane is found in the solution space to obtain the multi-view feature projection matrix which is sensitive to image segmentation.The main work of this paper includes:(1)The current image segmentation algorithms and semantic hashing algorithms are reviewed,and the existing problems are analyzed and found.(2)For two types of complex structure images,a pixel-level semantic similarity measure function is proposed,and a segmentation algorithm based on the semantic hash model is proposed and implemented.The validity of the proposed method is verified by comparing the simulation time and the evaluation index.(3)After extracting multi-view features such as space,color,saliency,and depth,we projected the features through the potential problem-solving plane to get low-dimensional features,and proposed a multi-view feature semantic image segmentation algorithm and realized it.Qualitative and quantitative comparisons of BSD500,Pascal VOC 2012 and other databases with the latest unsupervised segmentation algorithm and deep learning algorithm verify the superiority of this method.