Research On The Algorithm Of Color Digital Image Segmentation

Image segmentation is a basic computer vision technology,it is an important part of automatic image processing,it is the foundation of image analysis and understanding,and it has been attracting much attention.Since the 70 s of last century,researchers have proposed many methods for different images of different problems.By now,no specific algorithm is found suitable for most images ages,and there is no unified standard to choose the suitable segmentation algorithm.In this paper,a number of different image segmentation algorithms are improved.Finally,the proposed algorithms were verified by different datasets,and compared to existing algorithms.The main work is as follows:In this paper,in order to solve the segmentation degradation phenomenon when the number of super pixels is low,we propose a novel color image segmentation algorithm based on GrabCut.Firstly,Simple Linear Iterative Clustering(SLIC)algorithm is used to cluster the image,and the RGB mean of each pixel block that was clustering is then applied to reconstruct the simplified Graph Cuts model.Then,Bayes classification is used to classify hyper pixels in the simplified Graph Cuts model,and SLIC algorithm is then applied to cluster the image again,and the GMM parameter estimation is performed.Finally,the minimum cut algorithm is used to obtain the optimal image segmentation model.Experiments using real natural scene images demonstrate the superior performance of our proposed method.An improved algorithm for remote-sensing image segmentation based on combining spatial neighborhood information with FCM algorithm is proposed.In the initial stage,spatial neighborhood information and noise information was fused into the FCM algorithm objective function.The extension condition of the constraint is extended by the Lagrange multiplier method.The entire iterative process is optimized,and the iteration time is shortened.In order to ensure the noise is insensitiveness and image detail is preserved,we calculate the neighborhood-window data block instead of the single pixel.In the final stage,the FCM clustering is completed and the classification process of the image is accomplished.At this point,the final classification results were obtained.A new image segmentation algorithm based on bitmap cut and region merging is proposed.First,the spatial gradient operator is used to obtain the initial gradient image,bitmap cut is designed to reconstruct the gradient image during the subsequent applications.Then,the watershed segmentation is performed on the new gradient image.Finally,the regions of the presegmentation result are merged that based on the principle of minimum het-erogeneity,and the final segmentation result is obtained.The proposed algorithm has been tested on different images and compared with other existing algorithms,the experimental results show the proposed algorithm is effective and efficient.Particularly,for the more challenging fuzzy edge,the accuracy of the proposed algorithm outperforms other algorithms.A color image segmentation algorithm based on power law distribution is proposed,in order to solve the over-segmentation problem of the watershed algorithm.First,the original gradient image is obtained by the Soble operator,and the original gradient image is then calibrated by the minimum morphological value.Next,the watershed algorithm is used to segment the processed gradient image,and the area threshold is selected by the power law distribution relation between the pre-segmentation region area and the region number.Finally,the area threshold is introduced to destroy the self-organized criticality(SOC)state of the presegmentation label region.The region is then redistributed and the new SOC state is reached,thereby achieving the final segmentation result.Compared with the existing algorithms,this algorithm analyzes the region number-area distribution by means of statistical theory,and introduces the definition of SOC state.The new path of region merging is found by imitating the destruction the SOC state of the power law distribution,and a fast region merging of the pre-segmentation label regions is defined.The effect of noise on segmentation results is well suppressed,and a continuous,smooth segmentation line consistent with human vision can be obtained.