纹理感知相似性学习综述  

A Survey of Perceptual Texture Similarity Learning

作　　者：李传松 高颖 亓琳[2] 高峰 刘焱[3] 

机构地区：[1]国家税务总局青岛市税务局,山东 青岛 [2]中国海洋大学信息科学与工程学院,山东 青岛 [3]青岛酒店管理职业技术学院信息工程技术学院,山东 青岛

出　　处：《人工智能与机器人研究》2020年第1期8-15,共8页Artificial Intelligence and Robotics Research

摘　　要：作为物体表面的一种基本属性,纹理图像包含了纹理颜色、纹理基元等丰富的图像信息。在计算机视觉研究领域中,人们使用感知相似性来度量不同纹理之间的相似程度,研究人类对纹理图像的视觉感知。纹理相似性度量广泛应用于纹理识别和材质识别,是对象识别和场景理解的关键技术之一。可靠的感知相似性数据可以通过心理物理学实验获得,研究人员通过计算特征之间的距离度量估计纹理感知相似性。本文重点从纹理感知数据获取、纹理计算特征提取和纹理感知相似性估计三个方面回顾了纹理感知相似性学习的发展和常见的处理方法,并结合卷积神经网络对纹理感知相似性学习的对未来发展趋势作了分析。As a basic attribute of the surface of the object, the texture contains rich image information such as color and texton. People use perceptual similarity to measure the similarity between different textures. Texture similarity is widely used in texture recognition and material recognition, which is one of the key technologies of object recognition and scene understanding. Accurate prediction of texture perception similarity can help visual tasks such as texture retrieval and texture labeling to keep consistent with the results of human perception. Reliable similarity data can be obtained through psychophysical experiments, and researchers usually estimate the texture similarity by using the distance measure between the features of the textures. This paper focuses on texture perceptual data acquisition, texture computational feature extraction and texture perceptual si-milarity estimation, reviews the development of texture perceptual similarity learning and com-mon processing methods, and analyzes the future development of texture perceptual similarity learning combining with the convolutional neural networks.

关 键 词：纹理相似性 感知相似性 感知数据 特征提取 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]