Image Indexing Technique and Its Parallel Retrieval on PVM

The paper presents the technique for image indexing and its parallel implementation on a parallel virtual machine. The indexing method is the combination of the autocorrelogram method and color difference correlograms method. Color filtering is used for quantization of the RGB colorspace into 64 colors to reduce the complexity and calculation time. Then, both color and spatial information are employed to build the index of each image. Experimental results suggest that the method can index and retrieve images with different views of the same scene or large changes in appearance. The performance of the image retrieval system is also improved significantly when compared with both color histogram and color correlogram method. The scheme is also implemented on PVM with the load balancing technique to speed up its retrieval time. The experimental results also show that it could speed up the processing time significantly.