GPU Implementation of Edge Histogram Descriptor and Color Moments Fused Features for Efficient Image Retrieval
Abstract
In this paper, we implement the MPEG-7 Edge Histogram Descriptor and color moments fused features on the Graphics Processing Unit (GPU) to accelerate computations using CUDA. As the GPU can be used for data processing and deal with computationally enormous applications, in this study it is employed for efficient feature extraction phase of image retrieval. The Edge Histogram Descriptor describes the distribution of various types of edges with a histogram that can be a tool for image matching. These features are applied to search images from a database which are similar to a query image. We evaluated the accuracy of our method by the Precision-Recall graph. The average Precision and the average Recall of presented method are 71.53% and 55.00% respectively. GPU parallel computing led to a speedup of 15.12 over CPU sequential processing. Therefore, it demonstrates that parallel computing using a GPU can achieve good performance in image retrieval.
Keywords
Color Moments, Content Based Image Retrieval, Edge Histogram Descriptor, GPU, Parallel Computing