Segmented face approximation with adaptive region growing based on low-degree polynomial fitting

摘要: Nowadays, an objective in visual communication is to send and store images of faces at a low bit rate, such that the are still recognizable compression does not prevent remote face analysis. We present novel segmented approximation algorithm. Greyscale into meaningful surface segments with adaptive region growing algorithm based on low-degree polynomial fitting. The uses new thresholding technique \(L_\infty \) fitting cost as segmentation criterion. degree error automatically adapted during process. main novelty detects outliers edges, distinguishes between strong smooth intensity transitions, finds bent certain way, flat, planar, convex, concave or saddle patches. As result, correspond facial features, contours separating coincide real image edges. Moreover, curvature-based shape information facilitates many tasks automated analysis, demonstrated this paper by verification performed representation. representation provides good while preserving all necessary details reconstructed image. When compared different methods, we achieve higher ratios better rates. This confirmed correct identification percentages obtained recognition algorithms compressed data.