Saliency Region Detection Algorithm Based On Meanshift Clustering

When humans observe images,they can easily notice important parts of the target or interested images through the visual attention mechanism.The goal of significance detection is to achieve the same effect by computers.Significance detection can enable the computer to quickly extract the key areas from the image data,eliminate the interference factors in the image,and thus better process the image.However,detecting the significance regions of a given image accurately is a challenging task.Firstly,Significance objects have various characteristics that are difficult to be expressed by general mathematical models.Secondly,significance detection is usually used as a preprocessing part of other work,so significance detection should also meet the characteristics of speediness Significance detection is widely used in many fields such as automatic detection of targets,image retrieval,object recognition and image segmentation.By comparing some classical significance detection algorithms with image segmentation algorithms,this paper proposes a saliency detection algorithm based on mean shift clustering segmentation.Firstly,the algorithm is improved on the basis of FT frequency domain tuning algorithm.When the color contrast is weak and the salient area of the image is large,the salient image detection effect obtained by FT algorithm is not ideal.In this paper,BM3 D filtering is used instead of Gauss filtering in the original algorithm,which can smooth the edges of the image at the same time details,and then the image is transformed from RGB color space to Lab color space.The preliminary significance image is obtained by calculating the weighted Euclidean distance between the filtered image and the mean of the whole image.After obtaining the preliminary saliency map,a better segmentation region is obtained by improving the mean shift clustering segmentation method.Firstly,the original image is preprocessed before segmentation,and the noise of the image is removed.Secondly,the data sample set is continuously updated in the mean shift clustering process to speed up the calculation of the algorithm.Thirdly,after the initial image is segmented into regions by means of mean shift clustering,the algorithm is adopted.A certain merging strategy merges the regions,and then obtains the clustering segmentation image which is divided into several regions.Finally,the threshold value of each region is compared with the mean value of the preliminary saliency map obtained by the improved frequency domain tuning algorithm.The non-salient region is removed and the salient region is retained.The method proposed in this paper is compared with several other significance detection algorithms on the open data set ASD.The results show that the accuracy,recall rate and f-Meansure index of this algorithm are higher,and the detection results are more complete and accurate.