Sei sulla pagina 1di 1

Assignment 3- Advanced Zonal Search Algorithms

Motion Estimation (ME) reduces the temporal redundancy. The Block-matching algorithms are used in ME that calculate the best matching block using the Sum of Absolute Differences (SAD) criteria and determine the Motion Vectors (MVs). Full search (FS) algorithm searches all possible macro blocks within the search window for the best match and is hence computationally intensive. Diamond search (DS) algorithm reduces the complexity and achieves equivalent visual quality as FS by dividing the search area into diamond shaped zones which contain fewer checking points. This pattern can help in the overall performance of the estimation as motion vectors are coded in a pattern more similar to a diamond in terms of bits. But it has poor performance in the case of large global motion. Advanced Diamond Zonal Search (ADZS) : Advanced Diamond Zonal Search (ADZS) is faster than DS in terms of number of checking points and total encoding time and has high PSNR. This algorithm uses center-biased property of motion vectors based on the property that most MVs are closer to the (0, 0) MV, which speeds up the searching process. The zones are partitioned considering the fact that the innermost zones usually have a higher probability of containing the actual motion vector solution and this reduces the actual number of zones examined. The SAD is used as the distortion measure for this algorithm. It uses a set of fixed thresholding parameters that essentially enable the early termination of the search. The median value of the MVs of the adjacent blocks at the left, top, and top-right from the current position is selected for a predictor. Motion Vector Field Adaptive Search Technique (MVFAST): MVFAST considers the MVs of spatially adjacent blocks and the (0,0) motion vector as initial predictors, in order to improve the DS algorithm by selecting the best candidate and then performing a modified two stage DS pattern using this MV as the center. The main drawback of this algorithm is, like DS, the algorithm is zero-biased and has fixed threshold parameter. Advanced predictive diamond zonal search (APDZS): The APDZS is essentially an extension of the ADZS and uses multiple initial predictors. APDZS improves on ADZS further, by combining ADZS with 1) Generalized predictor selection-which uses a set of linear or nonlinear combinations of MVs such as the mean, median or a weighted mean of all the MV predictor candidates of the macro-blocks on the left, top, and top-right etc and selects the best among these as the final predictor to be used as center. It also uses acceleration candidates (MVs of the collocated block from preceding frames) as another set for prediction. It increases the correlation and reduces entropy. 2) Adaptive thresholding parametersexploit correlations between temporally and spatially adjacent blocks for calculating the necessary thresholding parameters. APDZS achieves higher PSNR. It is more stable and relatively fast compared to DS and MVFAST. PMVFAST improves upon MVFAST in speed and quality with the generalized predictor selection and adaptive thresholding parameters. It uses median predictor and it gives comparable PSNR as APDZS and is much faster than APDZS.

Potrebbero piacerti anche