Summary: | 碩士 === 南台科技大學 === 資訊工程系 === 96 === Edge detection is a very important preprocess for image processing research, we can decide the information about position, contours or size of the object by the features of image edge that detected by edge detection, and approach of pattern recognition and information retrial. But in general the edge detection methods, are required to manually select the parameters of an ideal edge to detect the edge, this method is not efficient.
This study used special gray-level conversion and Crimmins Speckle Removal as pre-treatment, and use of the Contextual-Hopfield Neural Network methods to unsupervised-learning methods to solve the problem manually select the parameters. Finally the use of voting to select the ideal edge imaging tests to assess the parameters of this study without the accuracy. The experiment proved that the method of unsupervised detection results can be similar to the edge of the ideal results.
|