The Design And Implementation Of An Image Segmentation System Based On Semi-Supervised Learning

Image Segmentation is one typical issue of research areas like Image Processing and Analysis, which has been playing an important role in Computer Vision and Image Engineering. Image is segmented to be sub-domains or objects on the basis of the un-continuity and similarity of luminance value, according to the problems to be solved. As classification using few labeled samples has come true through the adoption of Semi-supervised Learning, and the cost of obtaining labeled samples is much higher than unlabeled ones, it is significant to research on Image Segmentation using few labeled samples.After discussing the state of development in Image Segmentation technology and Semi-Supervised Learning theory, the process of applying Linear Neighborhoods Propagation algorithm to two-class classification is analyzed in detailed, on the basis of which it is extended to multiclass classification. In the process of applying Linear Neighborhood Propagation to Image Segmentation, firstly the image is optimally grayed and smoothed, then the weight graph of image data is constructed by quadratic programming, on the basis of above, labels is propagated to get all the labels, after that the foreground of image is generated by synthesizing the label and source image. Finally, contrastive analysis upon the results of Linear Neighborhood Propagation and another method is done, which shows that Linear Neighborhood Propagation is superior in both the quality and the requirement of manpower.Taking the Image Segmentation system as combination of theory and practice, function modules are set on the basis of concretely analyzing the requirements of practical application, and system organization is designed on the basis of in-depth analysis on the characteristic of C/S model and B/S model. Then source codes are programmed on the basis of detailedly designing the function modules and logic structure of data. Finally, functional test are processed and the results shows that the system meets the requirements and all modules run ordinarily. Image Segmentation system proposes a good combination of algorithm and application, provides friend user interface, and supplies good experiment platform for following research about theories and algorithms.