Please use this identifier to cite or link to this item:
Title: Adaptive Novel Pattern Search Block Matching Algorithm for Video Compression
Researcher : Shinde, Tushar Shankar
Supervisor: Tiwari, Anil Kumar
Department: Center for Information Communication and Technology
Issue Date: May-2015
Citation: Shinde, Tushar Shankar. (2015). Adaptive Novel Pattern Search Block Matching Algorithm for Video Compression (Master's thesis). Indian Institute of Technology Jodhpur, Jodhpur.
Abstract: Video data requires huge storage space and hence large bandwidth for transmission. This leads us to need of video compression where redundancies present in video are removed. Strong temporal correlation exists between consecutive frames of video, which can be used to remove inter frame redundancies through motion estimation and motion compensation. Motion compensation process is computationally demanding, hence our motive is to reduce the computational complexity without any significant degradation in prediction quality. There are many fast block matching algorithms available in literature. To reduce the computational requirements, we propose Adaptive Novel Pattern Search algorithm in which the search center is adaptively found by comparing block matching criteria values for default (0,0) and adjacent blocks motion vector. This helps us to directly arrive in the minima region. Additionally, we propose new Wings-Diamond search patterns for horizontal and vertical motion videos. For inclined motion content, we suggest to use Inclined-Hexagonal search patterns. These patterns helped us for fast convergence. We performed experiments for different types of motion sequences like slow, medium, fast and directional motion content videos. Search parameters like block size, search range and threshold were varied for comparative analysis. Some widely used algorithms available in literature were implemented and their performance in terms of entropy, prediction quality and complexity was compared with the proposed algorithm. In general, it was observed that Full Search algorithm provides very high prediction quality with huge computational complexity. Adaptive Rood Pattern Search reduces the complexity but looses in terms of prediction quality. It was found that for slow and medium motion videos our proposed algorithm works in line with Full Search in terms of prediction quality and with comparatively very low computational complexity. For fast motion videos Adaptive Rood Pattern Search performs better than Full Search considering the tradeoff between prediction quality and complexity. Our proposed algorithm works in line with Adaptive Rood Pattern Search. For directional motion sequences proposed algorithm outperformed all the other algorithms.
Pagination: x, 46p.
Accession No.: TM00073
Appears in Collections:M. Tech. Theses

Files in This Item:
File Description SizeFormat 
TM00073.pdf3.99 MBAdobe PDFView/Open    Request a copy

Items in IIT Jodhpur Theses Repository are protected by copyright, with all rights reserved, unless otherwise indicated.