基于环状矩形分块的纹理图像检索

Texture Image Retrieval Based on Ring-Like Rectangular Blocks

提出一种环状矩形分块纹理检索算法。将图像分成环状矩形边缘区域、中心环状矩形区域和核心矩形区域。对最外层边缘区域不予处理,这可明显减少检索时间,且对检索效果影响很少。对每幅图像分别求出中心区域和核心区域基于灰度共生矩阵的能量、对比度和熵所构成的纹理特征向量,两幅图像中心区域纹理特征向量欧式距离和核心区域纹理特征向量的欧式距离之和决定这两幅图像相似性。而环状矩形具有圆形分块的旋转不变性但计算更简单。经实验验证。

A new method for texture image retrieval which is based on ring-like rectangular blocks is proposed. It divides into the image the ring-like rectangle fringing field, the central ring-like rectangle region and the core rectangle region. No treatment is for the outermost edge area, this can significantly reduce retrieval time and little influence on retrieval effect. The texture feature vector composed of GLCM-based energy, contrast and entropy is calculated for each image central area and core area. Sum of two region euclidean distance decides these two image similarity. Ring-like rectangular has rotation invariance, but calculations easier. Validated by experiments, for some categories it has better search results.