A survey of perceptual image processing methods

Perceptual approaches have been widely used in many areas of visual information processing. This paper presents an overview of perceptual based approaches for image enhancement, segmentation and coding. The paper also provides a brief review of image quality assessment (IQA) methods, which are used to evaluate the performance of visual information processing techniques. The intent of this paper is not to review all the relevant works that have appeared in the literature, but rather to focus on few topics that have been extensively researched and developed over the past few decades. The goal is to present a perspective as broad as possible on this actively evolving domain due to relevant advances in vision research and signal processing. Therefore, for each topic, we identify the main contributions of perceptual approaches and their limitations, and outline how perceptual vision has influenced current state-of-the-art techniques in image enhancement, segmentation, coding and visual information quality assessment.     

基于注意度的线激光视觉测量的三维形状检测

针对工件在产线上的摆置方位识别困难识别率低等问题,提出了一种基于注意度的线激光视觉测量的曲线形状识别方法。采用激光视觉三维测量技术提取工件表面的光迹信息,将工件的方位识别转换为曲线形状识别的问题。在匹配算法方面,定义了两种距离测度,分别作为粗匹配和精确匹配的误差评价函数;在识别准确性方面,定义了一种基于注意度的模板相似性测度的分类方法。实验结果表明,本文提出的方法识别率高,耗时短,对工件的平移和偏移不敏感;能满足工业生产实时检测要求,应用范围广。     

基于置信区域内多级动态层表达的类贯序蒙特卡洛视觉跟踪方法

视觉跟踪是智能监控、机器人和视觉导航等领域的核心技术.针对现有类贯序蒙特卡洛跟踪方法复杂度高、实时性差的问题,本文提出了一种融合置信区域内多级动态层表达的跟踪方法,采用更加可靠、有效的粒子模拟状态后验概率.该方法利用检测模块得到目标可能出现的置信区域,根据真实目标尺寸给出一种粒子采样策略,每个粒子代表一级动态层表达,并为每个粒子建立双层运动模型;构建Mean-Shift分块观测模型以引入空间和外观信息,同时根据子块的匹配程度计算粒子权值、检测目标遮挡状态并提出模型更新策略.在公开视频序列上同经典粒子滤波和Mean-Shift等算法的实验对比结果证明了本文算法具有较优的跟踪准确度和实时性.     

Retina-Inspired Artificial Synapses with Ultraviolet to Near-Infrared Broadband Responses for Energy-Efficient Neuromorphic Visual Systems

Neuromorphic visual system with image perception, memory, and preprocessing functions is expected to simulate basic features of the human retina. Organic optoelectronic synaptic transistors emulating biological synapses may be promising candidates for constructing neural morphological visual system. However, the sensing wavelength range of organic optoelectronic synaptic transistors usually limits their potential in artificial multispectral visual perception. Here, retina-inspired optoelectronic synaptic transistors that present broadband responses covering ultraviolet, visible, and near-infrared regions are demonstrated, which leverage the wide-range photoresponsive charge trapping layer and the heterostructure formed between PbS quantum dots and organic semiconductor. Simplified neuromorphic visual arrays are developed to simulate comprehensive image perception, memory, and preprocessing functions. Benefitting from the flexibility of the charge trapping and organic semiconductor layers, a flexible neuromorphic visual array can be fabricated, having an ultralow power consumption of 0.55 fJ per event under a low operating voltage of −0.01 V. More significantly, an accelerating image preprocessing effect can be observed in a wide wavelength range even beyond the perception range of the human visual system, due to the gate-adjustable synaptic plasticity. These devices are highly promising for implementing neuromorphic visual systems with broadband perception, increasing image processing efficiency, and promoting the development of artificial vision.     

视觉密码方案概述

视觉密码(VC)是一种用于分享秘密图像的加密方案,与传统的密码技术相比具有简单性、安全性、隐蔽性等优点.其加密是将图像编码为分享,将分享打印在透明胶片上或以数字形式存储;其解密过程是对分享进行叠加,然后通过人类的视觉能力来实现.本文介绍了视觉密码的基本概念和各种不同的视觉密码方案,以及每种方案中使用的技术.此外,对视觉...     

基于小波变换的多尺度SSIM算法

提出一种基于小波变换的多尺度结构相似度的图像质量评价(WWMS-SSIM)算法。首先对原始图像和失真图像分别进行小波变换,在对应的低频子带图像中比较亮度和相关度差异,在对应高频子带图像中比较对比度、相关度和结构差异,然后赋予不同子带图像不同的视觉加权值,最后得到归一化的图像质量评价值。通过对不同失真类型图像以及Live数据库图像的评价,证明了提出的算法(WWMS-SSIM)相对均方误差、峰值信噪比、结构相似度以及改进的结构相似度评价算法更符合人类的主观视觉感受。     

一种带有视觉关注先验的联合特征非监督式抠图

针对谱抠图中组件划分整体性不强和前背景组件区分缺乏引导信息的问题,提出了一种带有视觉关注先验的联合特征非监督式抠图方法。首先结合有序径向差分纹理算子,提取图像颜色纹理联合特征,降维并聚类得到初始组件;而后充分利用人类视觉“关注图像中心而忽略边界”这一先验,设计各组件中心先验、边界先验和显著性指标,推导组件的前景和背景概率,并据此概率先验引导最优图层组合得到初步抠图结果;最后结合视觉先验信息和拉普拉斯矩阵的局部特性构造全局优化函数得到优化的抠图结果。实验结果表明,本文算法所提取的前景具有更强的整体性和细节保持能力。     

一种基于多级空间视觉词典集体的图像分类方法

针对单一特征时存在提取的信息量不足,对图像内容描述比较片面,提出将传统的SIFT特征与KDES-G特征进行串行融合,生成一个联合向量作为新的特征向量.针对传统的视觉词典构造方法缺乏考虑视觉词汇在空间的分布特点,本文引入图像空间信息,提出了一种空间视觉词典的构造方法,先对图像进行空间金字塔划分,再把空间各子区域内的特征分别聚类,构建属于对应子空间区域的空间视觉词典.在图像表示阶段,图像各子区域内的特征基于其对应的空间视觉词典进行LLC稀疏编码,根据各子区域对图像贡献程度的不同,把编码后各子区域的特征向量赋予不同的权重加权处理,再连接形成最终的图像描述.最后,利用线性SVM进行图像分类,实验结果表明了本文方法的有效性和鲁棒性.     

基于视觉对比机制的红外弱小目标检测方法

为解决复杂背景下红外弱小目标检测精度低的问题,本文提出一种基于视觉对比机制的红外弱小目标检测方法,算法主要模拟了人眼对目标对比度敏感这一机制。首先利用8向梯度方程提取红外图像的梯度显著图并二值化处理;根据小目标的尺寸大小特征对梯度显著图进行优化处理,剔除孤立的噪声点和尺寸较大的背景梯度显著区域;利用视觉对比机制对优化后的显著图进行局部对比度计算,通过阈值处理剔除虚警目标,完成红外弱小目标检测。仿真实验表明,该算法在低信噪比情况下对红外弱小目标的检测率较高,且虚警率低,单帧检测时间较小。