LVQ-GVF Snakes Model for Object-Based Segmentation

In this paper, we propose an LVQ-GVF snakes model for object-based segmentation of image sequences. Our method extracts some visual features, i.e. spatial (pixel position), color (YUV) and edge (canny edge detection) features. First, an initial object of interest is created by human assistance on a reference frame. Then, we employ Learning Vector Quantization (LVQ) with spatial and color features to provide the approximate location of object boundary. Starting from the approximate object boundary detected by LVQ, we employ Gradient Vector Flow (GVF) snakes to create more accurate contour of the object of interest. In the experiments, we implement this method as the preprocessing step of the 3-D human face reconstruction. The experimental results demonstrate accurate boundary of the object of interest (human face), which is an important part in reconstructing 3-D human face. Keywords: object segmentation, object boundary, gradient vector flow snakes, learning vector quantization.

Nama : Amang Sudarsono, Mochamad Hariadi, And Takafumi Aoki
Email : amang@eepis-its.edu
Kategori : SPEECH & SIGNAL PROCESSING
Institusi : Electronic Engineering Polytechnic Institute of Surabaya (EEPIS), Institut Teknologi Sepuluh Nopember (ITS), Surabaya, Indonesia Graduate School of Information Sciences, Tohoku University, Aoba-ya