An Image Retrieval System Based on Multiple Features

• Main Authors: Lu Yi Jun, 呂益君
• Other Authors: Jim Z. C. Lai
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/49794066533146657908

碩士 === 逢甲大學 === 資訊工程學系 === 87 === In this paper we propose a method which using colors, shapes and spatial relationships to retrieve images. In recent years, color information is the commonly used feature for content-based image retrieval. One of the well-known solutions is to retrieve images by color histograms. The advantages of color histograms are computation efficiency and insensitivity to small changes in camera viewpoint. However, retrieving images by color histograms has a poor performance for large database. This is because images with the same color histograms may have different appearances. In our method, we use multiple features to overcome this problem and try to retrieve images according to their overall appearances. We use color connectivity vector (CCV) which describes the distribution of color components in image to improve the performance of color histograms. We also use region shape vector (RSV) to represent the shape information in image and to increase the precision of image retrieval. In spatial relationships, we use 2D string to describe the spatial relationships of images simply. Then user can choose from these three features to search required images in an image database. Using our method to retrieve images has following characteristics: Images that have the same color histograms can make a distinction; Shape information will increase the precision of searching results; The similarity between two images can be easily computed by vector representation; The images in the database can be arbitrary.