Texture Image Segmentation Based On Markov Random Field Model

With the development of the computer technology, the collection and application of images are highly focused on and well developed. Image technique has wildly applications, such as the research, medical, education, communication. It has a great influence to the social and people. However, the contradiction between the huge image data and the inefficient image processing hinders the practical application of image techniques.Image segmentation is a fundamental step of image processing. Segmentation, target recognition, feature extraction and parameter measure transform the image to a compact form and provide the base for the image analysis and understanding. Because many images include plenty of texture information and precise description of the texture is crucial for segmentation, texture segmentation is one important part of image segmentation. Markov random field model (MRF) can well segment the texture images since it can suitably describe the spatial information and owns the prefect theoretical foundations. In this paper, we focus on the texture image and do the following works to improve the existed MRF model.(1) We introduce a region-based fuzzy multi-resolution MRF model. This model incorporate the region feature and the fuzzy theory to improve the multi-resolution MRF model by considering more statistical information. We demonstrate the validation of our model through the experiments for the synthesized texture images from the Brodatz database.(2) The wavelet transformation is usually used to describe the multiple resolutions of the image for the multi-resolution MRF model. However, wavelet transformation is a linear transformation, which can not effectively describe the non-linear features, such as the shape, size, contexture. We extend the multi-resolution MRF model form the wavelet domain to the morphological wavelet domain for the texture image segmentation. The experiments of texture images segmentation perform the validation of our model, where the test texture images are employed from the Brodatz and Prague texture image databases.