Research Of Mathematical Morphology And Random Field On Image Segmentation

Image Recognition and object detection have been the important techniques in image processing. Feature extraction and image segmentation are the most important parts in a recognition or detection system. Both of them have a great prospect in areas of theory and application. Then, we pay attention to these aspects, and propose some new algorithms.In feature extraction, we study on the morphological extracting methods, especially the computational complexity, robustness, and feature attributes, and propose morphological detectors based on points element and symmetric disk element. We make contributions in simplifying morphological operation, reinforcing feature representation, and equality of different structure elements. Some of the conclusions are widespread not only in feature extraction but also in other morphological methods. Our conclusions in feature extraction have a good performance in binary image processing and real sense image processing, which are verified by the theoretical analysis and experiments.In object detection, we employ Markov Random Fields in car detection, and propose a method based on non-isotropy random field in fast extracting car features and recognizing car bodies. Different from other detectors with classifier, the new method combines geometric features with Markov Random Field, reduces the complexity, and improves the accuracy of the detection.In image segmentation, we propose some methods based on the revised class adaptive spatial variant mixture model (CASVFMM). We discuss the convergence of CASVFMM model, analyze the reason for the method sensitivity to the initial value, and propose a new spatial variant mixture model. The segmentation method based on the new model has better performance in convergence and flexibility. It can alter the smoothness of different segment areas according to the types of input image and segment task, which has more appeal in real application. Then, we propose a segment model with non-homogenous Markov Random Field. The method based on this model can automatically estimate parameters for different segment areas respectively, which can benefit the task of multi-object segmentation. Based on the model of area parameters estimation, we study on the estimation method with morphological operation in Markov Random Field. With the shape analysis ability of morphological operation, we correct parameters estimation and the image segment result by the intermediate information from iteration. This method has a better performance in special area segmentation and analysis.We prove the efficiency of the new methods by the theoretical analysis. Great deals of experiment results from the standard test images verify the reality of them.