Algorithm Research And Implementation On Image Segmentation Based On Classified Image

Image segmentation is an important and basic problem in both image processing and computer vision. It is also a precondition of image analysis. On the one hand, image segmentation is the basis of object expression, and has a great influence on feature measure. On the other hand, it makes further image analysis and comprehension become possible, since original images can be transformed into more compact forms by using image segmentation, object expression, feature extraction, parameter measure and so on. That the result of segmentation is better or not has a strong influence on the following recognition and interpretation. Therefore, a lot of segmentation methods have been proposed every year. However, a universal method doesn't exist until now.The article expresses a research firstly to the current status of image segmentation and its practicability, and then brings forward a new image segmentation method, which mainly includes following aspects.(1) At the first chapter, the author introduces some classical theories on image segmentation, and those theories are foundations of following work.(2) An image segmentation algorithm based on texture and non-texture is proposed in this chapter. Better segmentation effects are achieved after the classified processing. Firstly, perform wavelet transform on images. Secondly, perform k-means algorithm in the LL sub-band of the images according to the color and texture features. Thirdly, judge that the images are texture or non-texture by using the region segmentation. The result of classification is a satisfaction.(3) To texture images, a novel approach based on wavelet-transform and using feature weighting is proposed in this chapter, which is to improve the accuracy of boundary locations and region homogeneity as well as to reduce the error rate in texture image segmentation. This new technique contains four consecutive stages: feature extraction, clustering number working-out, pre-segmentation and post-segmentation. In the feature extraction stage, texture features are extracted by using the pyramid-structured wavelet transform. The original image is then segmented initially using k-means algorithm in the pre-segmentation stage. According to the pre-segmentation results, the extracted features are weighted and the pre-segmented image is further processed with a minimum distance classifier in the post-segmentation stage to finally get the segmented image. All technical points are clearly described and presented in detail. Some segmentation experiments with different texture images are performed to test the performance of the new technique and are also included. Compared with a typical traditional method, the presentapproach shows visible improvements both in diminishing segmentation error, and in increasing the precision of boundary and region's harmony.(4) To non-texture images, a new color image segmentation method is proposed in this chapter, which based on union probability density of hue, light and saturation. The performance of this proposed method is demonstrated on a variety of images, and the experimental results show that the proposed method is better than traditionary methods.(5) At the last chapter, there is a summation to the whole work, and make a prospect to image segmentation.