| dc.description.abstract | Image segmentation is an important technique for the applications of computer vision and pattern recognition systems. Many image segmentation methods have been proposed based on gray levels, color values, gradient information, texture properties, etc. A new method called active contours without edges has been proposed to detect objects in a given image. The method employed the level set formulation and the level set method has abilities to deal with the cusps, corners, and automatic topological changes. In this study, we improve the method for multi-spectral and high-noised image segmentation.
In the active contour method, we need to set parameters in the Mumford-Shah function. Different parameters will result in different segmentation results. Moreover, the model does not perform well if the parameters are far different. In this study, we propose a method to adaptively set parameters for image segmentation. The adaptive improvement is useful for automatic image segmentation; especially for high-noised images.
From the experimental results, we can find that our scheme is reliable and more efficient. We spend time on computing entropy, but the number of iteration times of curve evolving for low-noised images is reduced. On the other hand, we improve the correctness of contour extraction for high-noised images. | en_US |