| Summary: | 碩士 === 國立高雄第一科技大學 === 電子工程研究所 === 104 === Based on iterative linear interpolation, the resulting degree-2 polynomial has a graph representing the texture surface in a local area. Therefore, we regard the coefficients of the polynomial as local features and design a texture recognition system. The higher degree coefficients better represent texture details than lower degree ones. Therefore, the proposed system outperforms those methods where only first-order differences are considered. For a two-variable degree-2 polynomial, there are 9 coefficients namely C00~C22. The histograms of the 9 coefficients are encoded into binary features for similarity comparison between a query image and database images. In the experiments, we use Corel 1000 image database for texture recognition and compared our system with LTP, LBP, and LTrP. The average precision ratio (APR) falls in [38, 62.94] and average retrieval ration (ARR) falls in [6.3, 38], which are little bit smaller than LTP, LBP, and LTrP. However, for specific classes of images such as Africa, elephant, dinosaur, and flowers, the proposed system has much higher precision.
|
|---|
