Databases of color images have become increasingly important in recent years. The text-based retrieval of images in such databases is practical only if descriptive text annotations accompany each image. Creating such text descriptions is a labor-intensive process requiring human interpretation of each image. Except for relatively small, static collections of images, the cost of generating text annotations is prohibitive. The desire to avoid the use of text-based image descriptors has therefore led to the investigation of feature-based descriptors that can automatically be extracted from images and indexed in the database. The definitions of these features and the algorithms for their automated extraction from a given image are the foci of most of the current research into content-based image retrieval (CBIR).

This project focused on improving the computational efficiency of the well-known color set algorithm for content-based image retrieval. The color set concept is a useful and efficient approach to image indexing and query in a way that combines color and spatial information. By indexing relatively important regions based on both their color content and their spatial locations, the color set method allows rapid retrieval of images matching a specified color-spatial query.

Several refinements of the color set approach were developed in this project. First, the process of determining the relevant color combinations for color sets to be indexed was made approximately one to two orders of magnitude more efficient than the original algorithm. Second, a B-spline descriptor of region shape, size, and position was incorporated to supplement the original method's rectangular bounding box. Describing contours as closed B-spline curves provided an accurate and storage-efficient means of indexing region location and shape. During image query, the convex hull property of B-spline curves was exploited to enable efficient determination of region containment of a specific point. Finally, an idea for improving the computational efficiency of approximating region contours with periodic, quadratic B-spline curves was investigated.

Publications:

1. Phillip A. Mlsna and Jeffrey J. Rodriguez, "Gradient and Laplacian Edge Detection," in The Essential Guide to Image Processing, 2nd ed., Al Bovik, ed. San Diego, CA: Elsevier, 2009, Ch. 19, pp. 495-524 (book chapter).
   
   
2. Phillip A. Mlsna and Jeffrey J. Rodriguez, "Gradient and Laplacian Edge Detection," in Handbook of Image and Video Processing, 2nd ed., Al Bovik, ed. San Diego, CA: Academic Press, 2005, Sect. 4.13, pp. 535-553 (book chapter).
   
   
3. Phillip A. Mlsna and Jeffrey J. Rodriguez, "Gradient and Laplacian-Type Edge Detection," in Handbook of Image and Video Processing, Al Bovik, ed. San Diego, CA: Academic Press, 2000, pp. 415-32 (book chapter).
   
   
4. Phillip A. Mlsna and Jeffrey J. Rodriguez, "Efficient Indexing of Multi-Color Sets for Content-Based Image Retrieval," in Proc. 4th IEEE Southwest Symp. on Image Analysis and Interpretation, Austin, TX, April 2-4, 2000, pp. 116-120. [ PDF ]