Near-duplicate Video Fast Detection Based On Global And Local Features Fusion

In this paper,based on the analysis of the state-of-the-art of Near-Duplicate Keyframe(NDK)and Near-Duplicate Video(NDV)Detection,we further study the following research of Near-Duplicate Video Fast Detection Based on Global and Local Features Fusion and gain some achievements.Firstly,Gray-Scale Pyramid(GSP)is processed for improving the global feature of color histogram.By constructing the spatial pyramid of luminance and cutting scale,the algorithm enhances illumination and cutting scale transform robustness of global feature.The experiments show that the GSP is more robust than color histogram in detection of illumination transformed NDK and scale transformed NDK,while for the other transformation,the validity of them are almost the same.When the similarity distance threshold ?(28)2,the average recall of GSP can reach 95.2 percent.Secondly,an algorithm of Scale Invariant Feature Transform based on Sparse Coding(ScSIFT)is proposed to make up for the shortcomings of SIFT features' high dimension and low efficiency.In order to improve the algorithm's speed and efficiency,the SIFT feature is sparse represented and the sparse feature indexing structure is established.Experiments found that the ratio of matching features of ScSIFT and SIFT are similar while the algorithm of ScSIFT is faster than the SIFT algorithm by almost 52 percent.Last but not least,with the combinations of GSP,Sequential Ordinal Measures and ScSIFT,an algorithm of Near-Duplicate Video Detection Based on Global and Local Feature Fusion(GLF-NDVD)is proposed.This algorithm fuses the advantages of speediness of global features and high accuracy of local features.The contrast experiments show that the GLF-NDVD is more efficient than the traditional algorithms,which accuracy rate can reach 78.2 percent.