| Summary: | 碩士 === 中原大學 === 機械工程研究所 === 104 === Fuzzy-Clustering enhancement is the most comprehensive approach in contrast enhancement for image processing. However, in order to keep an accurate cluster, Fuzzy-Clustering requires a great amount of computational effort in its iterative clustering process. With the advancement of vision science systems and technology, the image size is continuously growing and Fuzzy-Clustering is becoming more difficult to do in real time. In this thesis, a method utilizing the Probability Density Function (PDF) is proposed to reduce the data dimension in identifying clusters. The advantage of using the PDF removes the need for iterations and trials in the clustering process. This method is also compared to the traditional with K-means clustering. The efficiency and cluster accuracy are then compared. In the image data processing, each pixel is identified in the L* a* b* color space instead of the RGB color space. An image entropy function is then defined. By identifying the maxima of the image entropy, it is claimed that the image enhancement procedure is made automated and at the same time adequate contrast enhancement is reliably achieved. The method is then compared and tested with the Histogram Equalization method (HE) and Multi-Scale Retinex with Color Restoration (MSRCR). The results show that the proposed method, in low contrast images, performs superior in terms of its contrast enhancement and is very promising for high-resolution images.
Keywords: Probability Density Function, image entropy, automatic image processing
|
|---|
