Spectral-based temporal-constraint estimation for semi-automatic video object segmentation

This paper presents an approach to estimate the constraints on semi-automatic video object segmentation. It is performed by the assumption that a motion vector space is pixels movement direction of current to subsequent frame. The motion vector value is calculated by applying the Block Matching Algorithm (BMA). Its result is added to pixels image coordinates affiliated to the constraint in current frame in order to create one in subsequent frame. Subsequently, constraints are applied as a companion of an input image for the objects extraction conducted by matting technique. After segmentation results evaluation, the error rate of matte extraction has high results, since the pixel constraints in subsequent frames is spreading and getting away from the object area. It is as a result of difference motion vector values in adjacent blocks. We create the adaptive block around user constraint in order to overcome this problem. Then, the motion vector value is computed by the Euclidean Distance between the current and subsequent frame based on the Hue angle, Saturation, and Value (HSV) color models. When this algorithm is applied to separate the objects on the frame, sequences are reducing error up to 63.60%.