線形予測モデルに基づくテキスチャ境界推定法の実験的検討

抄録

The principle that the true segmentation of an image consisting of more than one texture reginon corresponds to the division which gives the minimum of the Total Sum of Final Prediction Errors (TSFPE) holds if each texture can be described by a two-dimensional linear prediction model. This paper describes experimental studies on the effectiveness of an edge detection method for texture image based on this principle. Experiments show that estimated edge locations which minimize TSFPE correspond to the true edges precisely. Estimation error of edge location caused by statistical deviations can be reduced by applying wider processing area and combining results for two different prediction directions. Comparisons between this method and the simple minimization method of the total sum of squared prediction errors are also given, which show the superiority of the proposed method.