| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 88 === Image segmentation plays an important role in the application of computer vision, pattern recognition and video compression. In all image segmentation schemes, watershed algorithm is most suitable for the pre-processing of MPEG-4. In order to increase the speed of image seg-mentation and maintain exact result, we proposed an improved watershed algorithm to process a grayscale image quickly base on an immersion process. Lastly, we implement an improved processing element (PE) of watershed algorithm by FPGA.
We propose an improved watershed algorithm to decide the labels of 3x3 pixels in the extraction of a pixel from the sorted array. The proposed algorithm has the following improvements:
1. From the number of the pixels to be processed, the processing speed of the proposed algorithm is about 25% faster than the Vincent''s algo-rithm.
2. From watershed PE architecture design aspect, the data translation rate of the proposed algorithm is faster than that of the Vincent''s algo-rithm. This advantage will reduce the data throughput of PE and elevate the speed of transmitting data.
3. Considering the architecture complexity, the gate count of proposed PE is fewer than that of PE that only decides a pixel label in an ex-traction from sorted array.
|
|---|
