Abstract: The value of element in SUSAN initial edge response matrix represents the total number of pixels in USAN, both the statistical value and the distribution of elements in the response matrix reflect image features. If the statistical value is treated as a grayscale value, the distribution of elements in the response matrix may be regarded as textural features of a grayscale matrix. So the image retrieval algorithm by GLCM through transforming SUSAN edge response matrix to grayscale matrix is presented, which is to calculate SUSAN edge response matrix of image first, then to convert the response matrix to a grayscale matrix according certain mapping rules, and then to calculate GLCM of the grayscale matrix and its features, finally to retrieval by features. Experiments show that the recall and precision of this algorithm will become satisfied after the number of retrieval results reaches to a critical value, and it also reflects certain affine transformation, illumination change invariance and anti-noise robustness.

Key words: Smallest Univalue Segment Assimilating Nucleus（SUSAN）, Univalue Segment Assimilating Nucleus（USAN）, initial edge response, grayscale transform, Gray Level Co-occurrence Matrix（GLCM）, image retrieval

摘要： SUSAN初始边缘响应矩阵元素值对应USAN中像素点总数，该统计值与响应矩阵元素分布均体现图像特征。若将该统计值视为灰度值，则响应矩阵中的元素分布特征可以视为灰度矩阵中的纹理特征。提出SUSAN边缘响应值灰度化转共生矩阵检索算法，即先计算图像的SUSAN边缘响应矩阵，再按映射规则转换为灰度矩阵，然后计算灰度共生矩阵及其各特征描述子，最后进行特征检索。实验显示，该算法的查全率与查准率在检索结果数量达到某临界点之后较为满意，且体现一定的仿射变换、亮度变化等不变性与抗噪鲁棒性。

关键词: 最小吸收核同值区, 吸收核同值区, 初始边缘响应, 灰度化, 灰度共生矩阵, 图像检索