WSEAS Transactions on Computers
Print ISSN: 1109-2750, E-ISSN: 2224-2872
Volume 14, 2017
Spectral-Based Semi-Automatic Segmentation of Video Object Using Constraint Estimation
Author:
Abstract: Motion vector is acquired by calculating the conclusive dissimilarity of blocks in the current frame and search block on next frames. Block Matching Algorithm (BMA) is employed to obtain the motion vector value. The result is added to pixel coordinates in current frame correlated with user constraints. Next, the object segmentation process is performed by matting techniques, after constraint scribble automatically occupies the next frame. However, matte extraction reveals a high error rate value after evaluation of segmentation results, caused by motion vector calculation which is as the driving of constraint parameter conducted in entire block. As result, position of pixel scribble is extending and far from object expected when motion vector value is applied. To solve the problem, calculation of motion vector performance is only on the block directly correlated to pixels scribble. This research presents an approach estimating constraint on semi-automatic segmentation of video object and the aims is to estimate the constraint in driving position of pixels scribble, where in the object extraction in a single frame is done with image matting, while the temporal domain motion estimation algorithm performed by Exhaustive Search of the BMA, but it is not robust algorithms for motion estimation on the label (scribble). Thus, in this study improved with the ES algorithms are developing and applying adaptive block SAD (Sum of Absolute Difference) to determine the distance vector. At final, the motion vector value is used to move the label from current frame to next frame. The result reveals accuracy improvement of 71.19%.
Search Articles
Pages: 14-22
WSEAS Transactions on Computers, ISSN / E-ISSN: 1109-2750 / 2224-2872, Volume 14, 2017, Art. #2