Design and Implementation of Efficient Techniques in Content Based Video Retrieval using Feature Extraction

Abstract

newline Video down loading, sharing, storing increases day to day life. To newlineretrieve desire video from huge dataset is challenging task. Images uses low newlinelevel features such as shape, texture, color and shape whereas high level newlinefeatures such as temporal features available in video. Text based video newlineretrieval is not give exact result compared to content based video retrieval. newlineFor smaller dataset machine learning using different distance metric such as newlineChi- square, Correlation, histogram intersection and Hellinger distance newlinemetrics used to get desire video. Content based video retrieval is Local newlinedescriptors starting from a particular pixel and finding nearest path in newlineneighborhood. Global descriptors use in transform domain whereas local newlinedescriptor for spatial domain. For larger dataset deep learning using newlineVGGNet-16, Dense Net 121, Inception Res Net V2, Mobile V Net, Res Net newline101, and Xception Net models are used. newlineThe model supports incremental feedback-based learning which is newlinedesigned using a correlation feature engine. This engine utilizes a novel newlineaugmented correlation metric, which combines different distance metrics to newlinemeasures for continuous training set updates. Due to which, the model s newlineperformance is incrementally improved after every iterative batch evaluation. newlineThe proposed model was tested on UCF101, Open Video, FIVR, Media Graph, newlineIVP, Columbia University Video, and HMDB human video datasets. CVRAD2 newlinealgorithm is more scalable and give accurate result. The design and performance newlineevaluation are analyzed in anaconda3 with python tool. Three parameters are newlineevaluated such as Accuracy, precision and recall. The proposed CVRAD2 model uses newlinea combination of effective dataset clustering, with ensemble deep neural network newlineclassifiers and augmented distance metrics to improve efficiency of CBVR for multiple newlinedatasets. This model also uses a combination of IFL and IDL in order to incrementally newlineimprove its performance w.r.t. number of evaluated video samples. Due to this newlinecombination, the proposed CVRAD2 model can accomp