الفهرس 
Text Based Image RetrievalOnly 14 pages are availabe for public view
 |  | from | 136 |  |  |
| | | from | 136 | | |
Abstract
Content based image retrieval is a challenging method of capturing relevant images from large databases. Although this area has been explored for decades, no technique has achieved the accuracy of human visual perception in distinguishing images. Whatever the size and content of the image database is, a human being can easily recognize images of the same category. In this thesis, the Content Based Image Retrieval (CBIR) system is first evaluated by considering color, texture and shape as visual features for describing the content of the images. Then, three different algorithms are developed to enhance the CBIR system performance. In the CBIR system evaluation, a comparative study is done considering the most common color features techniques so as to determine the most efficient features that increase the system effectiveness. Also, the effect of adding Gaussian noise or Salt & pepper noise on the retrieval accuracy of the CBIR system is studied in order to select the most appropriate filters. Also, the Discrete Wavelet Transform (DWT) decomposition is evaluated to determine the appropriate level and band that can work effectively with the CBIR system. After evaluating the CBIR system, three different algorithms are developed to improve both the retrieval accuracy and the retrieval time of the CBIR system. The first algorithm is developed based on both the Gray Level Co-occurrence Matrix (GLCM) and the DWT in order to increase the CBIR system performance. The second algorithm is developed by combining the most efficient color with texture features that extracted from the previous system evaluation. The third algorithm is developed by combining the most efficient color and shape features. The proposed algorithms are evaluated and compared with the most recent algorithms by using 1000 WANG COREL color image database. The results demonstrate that each type of the image features is effective for a particular type of images according to its semantic contents. But, by using a combination of these features improves the retrieval results for almost semantic image classes. Also, the experimental results indicate that the proposed algorithms increase the retrieval accuracy and decrease the retrieval time in comparing with the existing systems.