A novel parallel feature extraction method using HGAPSO and GLCO based SFTA

Abstract

Content based visual information retrieval system (CBVIR) is an important system to know the information of the images. Image is much more powerful than a document because it can say a lot more than a document itself. Feature extraction is one of the major steps of CBVIR system. For image classification, there are bunch of methods.Segmentation Based Fractal Texture Analysis (SFTA) is an efficient texture feature method among them for its higher precision and accuracy. For large number of Dataset, it is necessary for optimizing the feature extraction time and accuracy. As a result, we bring a new approach on SFTA algorithm on our research. We use an optimum multilevel thresholding hybrid method of Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), called HGAPSO with our proposed GLCO (Grey Level Classification Based Optimization) method for increasing the effectiveness of the SFTA technique. To avoid the computational complexity we have implemented our proposed HGAPSO based SFTA algorithm on NVIDIA Graphics Processing Unit (GPU).GeForce GTX 610 is fully utilized to perform the scanning to see its efficiency. Our experimental results show average 95.5% classification accuracy for our tested dataset and also the GPU based implementation experiences 120+ X speedup over conventional CPU implementation.