SPECTRAL-BASED VIDEO OBJECT SEGMENTATION USING ALPHA MATTING AND BACKGROUND SUBTRACTION

The main objective of this study is to separate object from video sequences. To separate the object from the video data is performed by combine several algorithms. The first stages, the video data is split into several frames, and the initial frame is treated as a keyframe. Object extraction on the keyframe require human intervention, namely by giving scribble on the regions of foreground and background. Matting technique uses a closed-form solution method applied in this stage. Further, the results used as a reference for object extraction in subsequent frames. To get the labels on the next frames, background subtraction algorithm is applied, and the result is used as the input image on the next frames. So that the object extraction in subsequent frames can be performed repeatedly using matting techniques. The experimental results show that the object extraction at the initial frame shows good results. However, the accuracy decreases when the object moves too fast and suddenly. Keyword : Segmentation, Alpha Matting, Background Subractio