Content-Based Image Retrieval Using Statistical Color Occurrence Feature on Multiresolution Dataset

摘要

In modern life, the increasing use of different image-taking devices made image acquisition no longer a difficult task. To access a huge quantity of images having different resolutions stored in the dataset, the images must be kept in an organized manner. Content-Based Image Retrieval (CBIR) is an application of image retrieval problem, that is searching for a digital image from image dataset. Then term "content" in the context refers to some features that can be derived from the image itself. Color features are one of the important content of image which plays a vital role in image retrieval. Existing color features concentrate the only occurrence of pixel values or the correlation between pixel values. This paper proposed a new color feature which combines information about color shade percentage, color pixel occurrence percentage, pixel having maximum and minimum occurrence altogether. At the same time, proposed feature vector has a significantly reduced length which reduce computational cost of the retrieval system. This new feature is applied to one computer-generated image dataset (NITW-7500) and it's translated and multiresolution version (using bilinear interpolation) and one standard natural image dataset (Corel- 1K). Performance is improved in all variations of computer-generated images.