Research On Color Image Segmentation Algorithm Based On Curvelet Transform

Image segmentation is a critical step in image processing and image analysis. The purpose of segmentation is to extract meaningful image of region. The definition of the image segmentation is that, how to extract consistency region and the interesting region during the process of image. Extraction accuracy of the target region will directly affect to the subsequent process. It is very important. Because the image will be affected by external factors, such as illumination changes, shadows and background noise, existing image segmentation methods are not satisfactory. There are often over segmentation and under segmentation. This paper, curvelet transformation as the reseach background, mainly reseachs the JSEG(J measure based SEGmentation) image segmentation method and the Markov random field image segmentation method based on the multi-scale.The goal of this paper is improving the classic segmentation algorithm with curvelet transform. Compared with the old segmentation algorithm, extracted object region of the improved segmentation algorithms in this paper meet the regional consistency requirement better, and closer to manual segmentation. In this paper, two improved schemes are proposed. One is the improved JSEG algorithm, the other is improved image segmentation algorithm based on multi-scale Markov random field.Curvelet transform’s image enhancement result is superior than which by the wavelet transform’s. In order to reduce noise, this paper use curvelet transform to improve JSEG algorithm in color image. At the same time, the edge of the image and other important information will be enhanced. Experiment shows that, compared with JSEG algorithm, this algorithm overcomes over segmentation and under segmentation problems.The improvement of image segmentation algorithm based on multi-scale Markov random field uses curvelet to describe the characteristics of each scale. To improve the defects of segmentation algorithm, this method uses the improving clustering algorithm as the initial segmentation, and adds weight to balance the tag and the characteristic field of energy. At the end, do experiments for improved Markov random field image segmentation method based on the multi-scale with other segmentation methods by using BSD300pictures. The experiments include subjective analysis, quantitative analysis and analysis of anti-noise performance. Experiments show that, the regions extracted by the improved algorithm in this paper have more semantic features, and more like the manual segmentation results.