Image segmentation is one of the most important techniques in digital image processing. Through image segmentation, the characteristics of people interested can be extracted from images. Image segmentation is the foundation of further steps to image recognition, analysis and understanding. This dissertation makes researches on different segmentation methods for different applications. In the IC reverse analysis engineering, we need extract the line objects from the color images. For this purpose, this dissertation introduces a watershed-based post-processing algorithm. Original over-segmented image is made by watershed transform and classified probability of each region is obtained using fuzzy clustering method. Then we extract the neighboring regions in the line direction of each region based on edge and spatial characteristic. Finally the classified probability of each region is updated by its neighboring regions. Experimental results show that the segmentation gives very good extraction result. In texture segmentation, through the application of a wavelet-domain Hidden Markov Tree (HMT) model with localized parameters, I analyze the wavelet coefficients in different scales of different texture pattern and get localized parameters in different scales. Then we cluster these localized parameters and extract the HMT parameters features of different texture pattern. Using these features, the texture image is segmented in each scale. After merging the segmentation results in different scale, we can get the finally segmentation results. Parameters of the HMT model are determined without complex training process and reflect the local character of texture. This makes the segmentation results good. This dissertation is also studied in the aspect of video segmentation. I try to research a kind of fast motion segmentation algorithm which can be realized by hardware. Two kinds of algorithms are tested. On the one hand, a simplified translation and rotation model with 3 parameters is established and the global motion of each video frame is estimated. After global motion compensation, the foreground and background can be separated by motion detection. On the other hand, a motion segmentation algorithm with automatic moving target detection is given. The video frame is divided into several sets, different set have different motion model, which reflect the objects and background in video frame. Firstly, motion parameter estimation is done for each set. Secondly, these parameters are merged to decide how much objects in this video frame. Thirdly, segmentation of this frame is done using these parameters. Finally, segmentation result is refined using spatial information. Theory analysis and experimental results show that the algorithms proposed in this dissertation have respective advantages, such as innovation, accuracy, simplicity, applicability, etc. This dissertation explores a reasonable way in the application domain of image and video segmentation.
修改评论