基于多小波直方图的纹理图像检索

为了进一步提高小波直方图的检索性能,提出了一种多小波直方图算法. 该算法根据多小波分解的特点,从能量角度出发,通过子带组合、0/1量化等运算生成多小波直方图. 与单小波直方图相比, 多小波直方图在实数域中同时具有正交、对称、短紧支撑和高消失矩等特性, 并且具有计算量小、精度高等特点. 纹理图像检索的对比实验结果表明,多小波直方图的检索精度比单小波直方图提高了16%.

Multiwavelet Based Histogram for Texture Image Retrieval

In order to enhance the performance of the scalar wavelet histogram, a multiwavelet histogram algorithm was presented, in which the multiwavelet histogram is constructed by combining the multiwavelet channels and 0/1 quantization in terms of the characteristics of the multiwavelet decomposition from the view of energy. In comparison with the scalar wavelet histogram, the multiwavelet simultaneously possesses desirable properties, such as symmetry, orthogonality, shorter support for given approximation order, and high vanishing order. The method has advantages of less computation cost and higher retrieval rate. In the contrasted experiment for the texture image retrieval, results show that the retrieval efficiency is raised 16%.