融合全局和局部相位特征的彩色纹理图像检索-《计算机技术与发展》

文章信息/Info

Title:: Color Texture Image Retrieval of Fusing Global and Local Phase Feature

Author(s):: XU Jia; QU Huai-jing* ; ZHANG Zhi-sheng; WANG Ji-wei; WEI Ya-nan; ZHANG Han-yuan; School of Information & Electric Engineering,Shandong Jianzhu University,Jinan 250101,China

Keywords:: texture image retrieval; pattern of local gravitational force angle; global phase feature; local phase feature; color space

摘要:: 针对目前彩色纹理图像检索系统存在颜色特征提取不充分、纹理特征提取与匹配的有效性亟待提高等问题,提出一种新的基于全局和局部相位特征融合的彩色纹理图像检索方法。该方法首次将局部引力角模式( Pattern of LocalGravitational Force Angle,PLGFA) 作为局部相位描述符用于彩色纹理图像检索。在该方法中,首先在适合人类视觉系统的 HSV 颜色空间中采用非均匀方式量化三个颜色分量,并通过构建颜色直方图充分地提取颜色特征,同时在 Gabor 复变换域针对 V 分量数据进行相对相位建模而提取全局纹理相位特征;然后,利用 PLGFA 描述符在 RGB 空间的灰度图像中提取局部纹理相位特征;最后,将这三种特征通过闭式 K-L( Kullback-Leibler) 距离和改进版本的 Manhattan 距离进行有机融合,从而实现彩色纹理图像的检索。在 VisTex( full / small) 、STex 和 ALOT 四个通用的彩色纹理图像库上得到的检索实验结果表明,综合考虑平均检索率和时间消耗两个性能评价指标,该方法较其他最新方法具有较明显优势,从而验证了其有效性和可行性。

Abstract:: Aiming at the problems of insufficient color feature extraction and urgent improvement of the effectiveness of texture feature extraction and matching,a new color texture image retrieval method based on global and local phase feature fusion is proposed. For the first time,PLGFA? ? ? ? ( Pattern of Local Gravitational Force Angle) is used as a local phase descriptor for color texture image retrieval. In thismethod,the three color components are quantized in a non-uniform way in HSV color space suitable for the human visual system,and the color features are fully extracted by constructing a color histogram. Meanwhile,the global texture phase features are extracted by relative phase modeling of V - component data in the Gabor complex transform domain. Then,the PLGFA descriptor is used to extract localtexture phase features from the grayscale image in RGB color space. Finally,the three features are organically fused with the closed-formKullback-Leibler ( K-L) distance and the improved Manhattan distance to achieve color texture image retrieval. Experimental results obtained on VisTex ( full / small) ,STex,and ALOT color texture image databases show that the proposed method has obvious advantages over the state-of-the-art methods considering average retrieval rate and time consumption,thus verifying its effectiveness and feasibility.