Region Query Image Retrieval System Based on Unicolor-Object Feature and Color-Complexity Feature

• Main Authors: Yi-Tung Liu, 劉溢桐
• Other Authors: Rung-Ching Chen
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/juay9y

碩士 === 朝陽科技大學 === 資訊管理系碩士班 === 91 === With the rapid advancement of digital image and Internet technology, tremendous amount of digital image data is created everyday. Therefore, developing a fast and accurate image retrieval system effectively to cope with the image data is necessary.Color histogram is one of the most commonly used image features among color-based image retrieving. The advantages of the color histogram include simple procedure and quick computation. Besides, this feature can resist the shift and rotation variants of the objects in images. However, the color histogram can only be used to describe the global characteristics rather than the local ones in an image, such as the color complexity of an image. Hence, this paper proposes three image features CDESSO (Color Differences on Edges in Spiral Scan Order), CVPIWO (Color Variances of the Pixels Within Identical Objects), and CVAAO (Color Variance Among Adjacent Objects) which can state the different style color complexities of an image, respectively.Generally, there exist two different kinds of image retrieval systems ── full color image retrieval system and region-based image retrieval system. In a full color image retrieval system, a user can employ the tool like a scanner to input a query image Q into the system. The system then compares the features of the query image with those of each database image, which have been extracted in advance and saved in database. The database images which are most similar to Q would be transmitted back to the user.In a region-based image retrieval system, a user selects a region image RQ from the query image; then the system submits to the user the database images each of which contains a region image that is similar to RQ. Based on the proposed image features above, this paper also provides some full and region-based image retrieval systems. In addition, this paper analyzes the properties and the resistances of the proposed systems, like the shift, rotation, scale, noise, distortion, hue, and lamination variants.