الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 105 |  |  |
| | | from | 105 | | |
Abstract
Image segmentation has an important role in many applications of computer vision and image analysis. Color image segmentation methods are proposed in several ways. In this thesis, we develop an automatic superpixel generation method firstly. In this method, the quick-shift parameters are modified automatically based on invariant images considerably the algorithm speed. The proposed method segments images into homogeneous portions or regions by applying quick-shift method with initial parameters and then changes quick-shift parameters values automatically to get the final segmented image. This method is valid for large size images. The automatic quick-shift method is applied at two levels to extract superpixels from the invariant images. The invariant method reduces specular highlight, shadow, and shading which affect the appearance and analysis of the color images seriously but exceed the processing time, so we divide image according to its size to reduce invariant time. A quantization process is applied on the invariant image to be used as a reference image. Instead of using a specific value, we changing parameters values in iterations that make the proposed algorithm robust and flexible against different image characteristics. Finally, we develop a block-based invariant representation for color images considering images includes metal materials and dielectrics. This method divides the image into sub-blocks and applies the invariant equations within each block. This method is applied to color image segmentation problem. The effectiveness of this method for a variety of images including different objects of dielectrics and metals are examined and has a clear and sharp edge over the pixel-based algorithm