This item is available under a Creative Commons License for non-commercial use only
2. ENGINEERING AND TECHNOLOGY
Block matching is the most computationally demanding aspect of the video encoding process. In many applications real-time video encoding is desired and therefore it is important that the encoding is fast. Also where handheld devices such as a PDA or mobile phone are concerned a less computationally intensive algorithm means a simpler processor can be used which saves on hardware costs and also extends battery life. An optimised algorithm also allows these devices to be used in low bandwidth wireless networks. The challenge is to decrease the computational load on the system without compromising the quality of the video stream too much, thus enabling easier and less expensive implementations of real-time encoding. This thesis appraises some of the principal Block Search Algorithms used in Video compression today. This work follows on from the work of Aroh Barjatya who implemented 7 common Block Search Algorithms to predict P-frames in MATLAB. Three further hybrid DS algorithms are implemented in MATLAB. Additional code is added to produce plots of the main metrics and to calculate some statistics such as Average Searching Points, Average PSNR and the Speed Improvement Ratio with respect to the Diamond Search and the Exhaustive Search. For a comparative analysis with previous studies 3 standard industry test sequences are used. The first sequence, Miss America is a typical videoconferencing scene with limited object motion and a stationary background. The second sequence, Flower Garden consists mainly of stationary objects, but with a fast camera panning motion. The third sequence, Football contains large local object motion. The performance of the 3 implemented algorithms were assessed by the aforementioned statistics. Simulation results showed that the NCDS was the fastest algorithm amongst the 3 hybrid DS algorithms simulated. A speedup ranging from 10% for the complex motion sequence Flower Garden to nearly 54% for the low motion video conferencing sequence Miss America was recorded. All 3 algorithms performed very competitively in terms of PSNR compared to the DS even though they use a lower number of search points on average. It was shown that the NCDS has marginally worse PSNR performance than the DS compared to the other 2 algorithms – the highest being a drop in PSNR of 0.680dB for the Flower Garden sequence. However, the speed improvements for NCDS are quite substantial and thus would justify its use over the DS. The results from the implementation concurred with the literature therefore validating the implementation. The implementation was used as a guide in nominating a ‘robust’ Block Search Algorithm. When the DS, CDS, SCDS and the NCDS were compared with ARPS it was shown that ARPS generally gave both higher PSNR and higher search speed for all 3 sequences. The reason for the good performance of ARPS is that it quickly directs the search into the local region of the global minimum by calculating the Predicted Motion Vector. The minimum error from a rood pattern of nodes is found and then a final refined search calculates the motion vector. Simulation results showed that ARPS was the best algorithm amongst the 10 algorithms simulated from the point of view of speed (lowest number of search points used per macroblock) and video quality (PSNR). For real-time encoding of video the best fast block motion algorithm to advise is ARPS.
Casey, J.: An Investigation of Block Searching Algorithms for Video Frame Codecs. Master's Dissertation. Dublin, Dublin Institute of Technology, 2008