Research On Image Segmentation Application Based On The Normal Cloud Model

Image is the largest and most intuitive information carrier in people’s daily life. The image processing and analysis problem has become more and more important.Image segmentation is the pivotal step during the processing of image processing. It is the basis of various advanced visual information processing. In the field of image segmentation problems, one of these problems is the indeterminacy of pixel value of image. The traditional method of fuzzy theory and probability has some disadvantage in solving this problem. Cloud model is based on both of fuzzy and probability theory so that it can solve the problem of concept separation in the procedure of image processing. The method of normal cloud model segmentation algorithm has many peculiarities, such as simple algorithm model, exquisite segmentation effect and comprehensive partition regions, etc. Based on the studying on the cloud transform and cloud merging process under the architecture of the normal cloud model, this dissertation achieved the following results by improving the key step of transform and merging:Propose the multi-kernel extract algorithm based on the single-kernel extract algorithm. The image segmentation based on normal cloud model need s to do cloud transformation of the original image. During the procedure of cloud transformation, the expectations of cloud models are determined by extracting cloud kernel. Thus, the result of segmentation is distinguishing because of different cloud kernel extraction methods. Based on this, the dissertation proposes a new method of Multi-Kernel Extraction(MKE) on the basis of Single-Kernel Extraction(SKE). The MKE algorithm can extract multiple maximum values of image’s gray level at the same time and realize the parallel extraction of cloud kernels. The experiment results show that the MKE algorithm run faster and can decompose more cloud models than the SKE one. That makes that the cloud combination can get more basic concepts of the original image and the final division of the image segmentation will be more accurate.Propose the maximum-similarity concept promotion algorithm by improving the method of choosing models for merging in concept promotion. The cloud models which are the results of cloud transformation need to be merged by cloud combination. The cloud combination is the procedure of concept promotion which is based on certain rules. In this dissertation, it proposes the Maximum-Similarity Concept Promotion(MSCP) strategy based on the Minimum-Distance Concept Promotion(MDCP) and prove the theoretical validity of MSCP through the mathematical formula. The experiment data show that the results of MSCP are more approximate with the gray level pro bability distribution of the original image, so that the merged curve of cloud model is closer to the original image and the various concepts represented by the image region are more qualified with people’s cognition.Considering the influence of cloud transformation and combination to the result of image segmentation, this dissertation will combine two kernel extraction methods and concept promotion method with each other. By judging the segmentation result of different algorithm from both subjective and objective side, it can be considered that the performance of MKE-MSCP and MKE-MDCP are better. Image segmentation experiment results show that the algorithms have a strong adaptability and better image segmentation effect evaluation coefficient of the experimental images.