基于颜色四通道及空间金字塔的鱼类图像检索

随着计算机视觉技术在海洋水产领域中的应用不断加深,鱼类图像检索在渔业资源调查、鱼类行为学分析等方面发挥了巨大的作用。通过研究发现,鱼类图像的背景信息会对鱼类图像检索造成极大干扰,而且鱼类图像中颜色、纹理、形状等特征由于空间位置信息的缺乏而使检索的准确率不高。为解决以上问题,提出了一种新的基于颜色四通道及空间金字塔的鱼类图像检索算法。首先,提取视觉显著性图将鱼类图像的前景和背景分开,从而减少图像背景对检索的干扰;其次,为了使图像特征包含一定的空间位置信息,利用空间金字塔的理论对图像进行分割,在此基础上,将图像转为HSVG四通道图并提取SURF特征;;最后,得到检索结果。为验证所提算法的有效性,在QUT_fish_data数据集和DLOU_fish_data数据集上对算法的查全率、查准率与经典的HSVG算法和显著性分块算法进行对比:在两个数据集上查准率分别比传统的HSVG算法最多分别提高12%和5%,查全率最多分别提高7%和22%;比传统的显著性分块算法查准率最多分别提高15%和5%,查全率最多分别提高36%和22%;从而证明所提算法是有效的,能有效提升鱼类图像的检索效果。

Fish image retrieval algorithm based on color four channels and spatial pyramid

With the development of the application of computer vision in the field of marine fisheries, fish image retrieval has played a huge role in fishery resource survey and fish behavior analysis. It is found that the background information of fish images can greatly interfere with fish image retrieval, and the fish image retrieval results only using color, texture, shape and other characteristics of fish images are not accurate due to the lack of spatial position information. To solve the above problems, a novel fish image retrieval algorithm based on HSVG (Hue, Saturation, Value, Gray) four-channel and spatial pyramid was proposed. Firstly, a visual saliency map was extracted to separate the foreground and the background, thereby reducing the interference of the image background on the retrieval. Then, in order to contain certain spatial position information, the fish image was converted into an HSVG four-channel map, and on this basis, the theory of spatial pyramid was used to segment the image and extract the SURF (Speed Up Robust Feature). Finally, the search results were obtained. In order to verify the effectiveness of the proposed algorithm, the recall and precision of the algorithm were compared with classic HSVG algorithm and saliency block algorithm on QUT_fish_data dataset and DLOU_fish_data dataset. Compared with traditional HSVG algorithm, the precision on two datasets is increased at most by 12% and 5%, and the recall is increased at most by 7% and 22%, respectively. Compared with saliency block algorithm, the precision on two datasets is increased at most by 15% and 5%, and the recall is increased at most by 36% and 22%, respectively. So, the proposed algorithm is effective and can improve the retrieval results significantly.