Region-based Image Retrieval Using Multiple Features

  • Author(s) / Creator(s)
  • Technical report TR02-10. Large image databases are becoming popular due to the ease with which images are being created/digitized and stored. Content Based Image Retrieval (CBIR) has therefore evolved into a necessity. It is a challenging task to design an effective and efficient CBIR system. Current research works attempt to obtain the semantics or meaning of the image to perform better retrieval. Segmentation of an image into regions may reveal the true objects in the image and the local properties of regions can help matching objects between images and thereby contribute towards a more meaningful CBIR. This thesis proposes a CBIR algorithm, SNL, that utilizes the regional properties of the images. Each image is segmented and features including the color, shape, size and spatial position of each region is extracted. Regions are matched by comparing the region content, shape and spatial position and the IRM distance measure between the whole images is calculated. The relative importance of the above features are investigated. SNL outperforms Global Color Histograms and Color Based Clustering techniques in terms of precision-recall. The experiments were conducted on databases of size 10K, 20K and 50K images. Our representation is also robust to minor inaccuracies due to image segmentation. A more efficient version of SNL, SNL+, is designed using a filtering technique called the OMNI approach proposed in. The HF algorithm proposed in the paper is modified to select a better set of foci. The thesis shows taht using IRM, a non-metric distance, with the OMNI approach significantly reduces the query time without losing effectiveness. | TRID-ID TR02-10

  • Date created
  • Subjects / Keywords
  • Type of Item
  • DOI
  • License
    Attribution 3.0 International