Index image generation using DSP in high-definition still image retrieval

This paper describes the automatic generation of index images with the DSP in high definition still image retrieval and display systems. Index images generated from the high definition still images to be searched for have to be characterized by their features to a great extent. A Sobel filter was selected among the well known spatial edge filters, i.e., Prewitts, Sobel, and Laplacian, because the Sobel filter was suitable for picking up the object from an image with its background. Additional processing was carried out to make the contours of the object clear including the enhancement of their magnitude and gamma using the Sobel filter. The automatic index generation system was built using DSP applications. Each index image was automatically generated in 110 seconds whereas the index image generated in accordance with the conventional method took 35 minutes for complete processing. Samples of Kosode garments used in the study were prepared from digital archives of Kosode garments owned by the National Museum of Japanese History. The precision of indexing was 99.6% in accordance with the proposed automatic operations as compared with the conventional manual operations.