一种新的基于自适应提升小波的图像检索算法

Algorithm of Adaptive Lifting Scheme-based Texture Image Retrieval

基于小波域的图像检索算法主要存在两个问题:首先是噪声导致图像的不适当分解,降低特征的提取质量,影响检索结果;其次是从分解后的子带图像中提取的梯度信息,作为描述特征之一,对于图像的几何变换,特别是旋转和镜像非常敏感.针对上述问题,本文提出一种新的选择子—梯度幅值描述子对图像进行自适应分解,降低了噪声对检索结果的影响,并设计了一个名为合成排序梯度直方图的特征向量以便更好地衡量图像相似性.实验证明,本文算法具有良好的缩放、旋转和镜像不变性,检索结果能较好地符合人的视觉感受.

There are two disadvantages in the image retrieval based on wavelet transforms. The one is that unsuitable decomposition of image leaded by noise may influence the retrieval result, the other is the gradient information extracted from sub-images, as one of features, is over sensitive to geometry transforms. In this paper, a novel algorithm for image retrieval using a new descriptor named Gradient Amplitude Descriptor （GAD）, which aims at improving the adaptive wavelet lifting scheme based on Neville filters is proposed. To measure the geometrical and visual similarity between images, the Synthetic Sorted Gradient Direction Histogram （SSGDH） is generated, as a new feature vector, by gradient calculation of wavelet coefficients. Comparisons with prior methods show superiority in achieving both robustness to geometrical variations and optimum acutance matching to human visual perception.