自适应多尺度分块压缩感知算法

目的 基于小波域的多尺度分块压缩感知重构算法忽略了高频信号在重构过程中的作用,丢失了大量的边缘与细节信息。针对上述问题,提出一种自适应多尺度分块压缩感知算法,不仅合理利用低频信息还充分利用图像的高频信息,在图像细节复杂度提高的情况下保证图像重构质量的提高。方法 首先进行3层小波变换,得到一个低频信号和9个高频信号,分别进行小波逆变换后分成大小相同互不重叠的块,对低频部分采用2维邻块边缘自适应加权滤波的方法进行处理,对高频部分采用纹理自适应分块采样,最后利用平滑投影Landweber（SPL）算法对其进行重构。结果 与已有的分块压缩感知算法、基于边缘和方向的分块压缩感知算法和基于纹理和方向的分块压缩感知算法相比,本文算法在不同的采样率下,性能均有所提升,代表细节信息的高频信号得到充分重建,改进的算法所得到的重建图像具有较高的分辨率,尤其对细节较为丰富的图像进行重建后具有较高的峰值信噪比;2维邻块边缘自适应加权滤波有效的去除了重建图像的块效应,且重建时间平均减少了0.3 s。结论 将三层小波变换后的高频分量作为纹理部分,利用自适应多尺度分块重建出图像的轮廓与边缘;将低频分量直接视为平坦部分,邻块边缘自适应加权滤波重建出图像细节,不仅充分利用了图像的高低频信息,还减少了平坦块检测过程,使得重建时间有效缩短。经实验验证,本文算法重建图像质量较好,尤其是对复杂图像明显消除了块效应,边缘和纹理细节较清晰。因此主要适用于纹理细节较复杂的人脸图像、建筑图像和遥感图像等。     

基于模糊增强的图像区域分割算法研究

针对图像平坦区、纹理区和清晰边缘的分割问题,提出了一种基于模糊增强的图像分割算法.该算法依据基于模糊增强的Canny边缘检测原理,在充分分析图像纹理区和清晰边缘的像素分布特点的基础上,通过增强纹理区像素对比度,检测出更多的纹理区细节.并利用膨胀、区域连通等方法实现了图像的区域分割.实验结果表明,该算法能够准确地实现了图像平坦区、纹理区和清晰边缘的分割,并有较强的抗噪能力.图像分割结果可以反映更多的纹理细节信息.     

融合NSCT和自适应平滑的光照不变量提取算法

为了更好地解决光照变化对人脸识别系统的干扰问题,提出一种融合了NSCT和自适应平滑的算法,以提取带有更多人脸结构信息的光照不变量.首先用NSCT分解对数域人脸图像,并对各高频子带进行NormalShrink阈值滤波;再将滤波后的高频子带和未经处理的低频子带进行逆NSCT处理得到人脸图像的模糊图像;然后对NSCT分解后的低频子带使用自适应平滑提取出低频子带中的人脸细节信息;最后结合该人脸细节信息和模糊图像进行计算,得到人脸图像的光照不变量.该不变量有效地弥补了NSCT方法中缺乏低频子带中的人脸细节信息的不足,提高了人脸信息的利用率.在Yale B和CMU PIE人脸库上的实验结果表明,该算法能够有效地消除光照变化的影响,具有更优的人脸识别性能,提高人脸识别系统的光照鲁棒性.     

基于PCA与残差补偿的人脸超分辨率技术

针对传统人脸图像超分辨率算法中存在高频细节信息损失过多的问题,提出一种基于主成分分析和位置块残差补偿相结合的人脸图像超分辨率算法.首先使用主成分分析得到初步的人脸图像分块,再使用基于位置块的残差补偿算法,对初步的人脸图像分块进行高频细节补偿得到最终结果.实验结果验证了算法的有效性.     

融合先验信息的残差空间注意力人脸超分辨率重建模型

针对公共场所监控图像中低分辨率人脸图像利用现有人脸识别系统识别准确率低的问题,提出了融合先验信息的残差空间注意力人脸超分辨率重建模型,用该模型对低分辨率人脸图像进行预处理后再进行识别可大大提升识别准确率.该模型将面部先验结构信息嵌入到生成对抗网络模型中,再采用残差空间注意力激活算法突出空间位置中携带高频信息的特征,最后使用多阶特征融合算法充分利用不同尺度的特征,防止携带高频信息的人脸特征在网络传播中丢失.实验结果表明,重建出的超分辨率人脸图像具有更多的面部细节特征,大大提高了对低分辨率人脸图像的识别准确率,并且与其他5种模型相比,新模型具有较低的耗时和较少的参数.     

改进的非局部均值视频超分辨率重建算法

针对基于非局部均值(NLM)的视频超分辨率重建方法存在结果过于平滑、收敛速度慢及计算量大等问题,提出一种改进的NLM视频超分辨率重建算法。该方法采用模糊边缘补足算法将经过预处理的视频图像分成平坦区和纹理区;对于平坦区,采用直方图均衡化的方式进行图像增强处理,以减少算法计算量;对于纹理区,采用改进的NLM重建算法进行处理,通过设计多方向自适应搜索窗并引入邻域相干系数修正相似性权值,以增强重建图像的纹理细节,加快算法的收敛速度;将重建的纹理区与增强的平坦区进行叠加归一化处理,完成整个视频图像的超分辨率重建。实验结果表明,该算法能够在提高重建图像纹理细节及峰值信噪比的同时,降低算法的整体复杂度,缩短重建时间。     

结合图像细节特征的全局优化纹理合成

提出了一种结合图像细节特征的全局优化纹理合成算法.作者利用非线性分解方法获取纹理细节作为特征图,将其作为纹理图像一个新的信息通道来引导纹理合成.联合图像细节信息通道和颜色通道,在图像上建立了一个全局的纹理合成能量方程,采用最大期望值算法迭代优化求解出一个全局的能量最小值.在全局优化算法中,构建样本纹理和待合成纹理之间的各个信息通道直方图匹配算法,进一步提高纹理合成的质量.文中同时给出了结合细节特征的图像和视频纹理合成算法,实验结果表明,文中的方法较好地解决了纹理合成中合成的纹理图像结构不连续或特征结构容易断裂等问题.     

基于感知去模糊的高效人脸图像修复算法EI北大核心CSCD

针对现有深度学习图像修复算法在修复区域内部产生模糊纹理或存在修复区域边界的内容不连贯性的问题,提出一种基于感知去模糊的人脸图像修复算法.首先提出一种包含PSNR损失、SSIM损失和对抗损失的生成对抗网络生成粗略的人脸结构修复信息;然后开发一种基于感知去模糊的生成对抗网络对粗略的人脸结构进行进一步的纹理细节恢复,从而生成自然清晰且内容连贯的面部纹理.在CelebAHQ公开人脸数据集上进行定量、定性和消融实验的结果表明,所提算法在峰值信噪比和结构相似性等定性评价指标方面优于所对比的前沿图像修复算法.     

基于轮廓模板和自学习的图像纹理增强超采样算法

提出一种以轮廓模板插值和局部自学习相结合的图像纹理增强超采样算法,有效地恢复了插值图像丢失的细节纹理,抑制了插值图像边缘的扩散.该方法通过局部自相似性在原始低分辨图像中估计高频信息,对轮廓模板插值图像的细节纹理进行了恢复.其中,为了弥补轮廓模板插值缺少先验知识的缺陷,将原始低分辨率图像的高频信息作为先验知识.为了保证估计的高频信息最优,匹配的过程中采用双匹配,相比较于全局搜索和小窗搜索,提高了效率并保证了匹配精度.此外,使用高斯模糊代替了传统提取高频信息的方法,简化了算法的复杂度,提高了准确性和效率.对估计得到的高频信息采用高斯函数加窗,以减小估计出错和重叠区的混叠影响.本文算法的训练库由原始低分辨图像自身和插值图像构成,节省了生成训练库所需的时间和空间.训练库的简化使得高频信息的估计可以多尺度进行,算法效率得到进一步优化.理论分析和实验结果表明,相比传统的基于插值、基于自学习的图像超分辨率方法,本文方法获得更好的实验结果,主观效果得到明显改善,有效地恢复了图像的纹理细节,提高了图像边缘锐度,避免了产生锯齿等人工效应,客观指标得到提高.     

肤色纹理保留实时人脸美化算法

针对目前人脸美化算法容易产生皮肤纹理信息损失,眼睛和头发等区域信息丢失等缺陷,提出一种具有肤色纹理保留性能实时人脸美化算法.首先根据自适应肤色椭圆模型处理色度分量估计肤色相似度图;然后对原图局部方差处理得到肤色粗糙度,据此调整肤色纹理保留平滑参数,得到肤色纹理保留平滑图像;再利用肤色相似度图融合原图与肤色纹理保留平滑图像;最后对融合图像进行美白处理,获得人脸美化图像.在公用人脸数据库上的测试结果表明,文中算法不仅有效地祛除皮肤瑕疵,还能保留人眼敏感的肤质细节,使视觉感知比较自然;与现有美化算法对比,该算法美化效果更佳,鲁棒性强,复杂度低,能实时处理人脸视频.     

Hallucinating Compressed Face Images

A face hallucination algorithm is proposed to generate high-resolution images from JPEG compressed low-resolution inputs by decomposing a deblocked face image into structural regions such as facial components and non-structural regions like the background. For structural regions, landmarks are used to retrieve adequate high-resolution component exemplars in a large dataset based on the estimated head pose and illumination condition. For non-structural regions, an efficient generic super resolution algorithm is applied to generate high-resolution counterparts. Two sets of gradient maps extracted from these two regions are combined to guide an optimization process of generating the hallucination image. Numerous experimental results demonstrate that the proposed algorithm performs favorably against the state-of-the-art hallucination methods on JPEG compressed face images with different poses, expressions, and illumination conditions.     

Face segmentation and tracking based on connected operators and partition projection

A new technique for segmenting and tracking human faces in video sequences is presented. The algorithm relies on the concepts of connected operators and partition projection to tackle the problems of face segmentation and tracking, respectively. It uses a connected operator to extract the connected component that more likely belongs to a face. Such a connected operator is implemented by means of a Binary Partition Tree. A set of connected regions (a node in the tree) is selected maximizing the estimation of the likelihood of being part of a face. This way, a first estimate of the face is obtained. Final face segmentation is carried out by a face extraction step that applies the previous maximization in the partition space. Faces are tracked through the sequence based on the partition projection approach. Partition projection accommodates the partition of the previous image into the information of the current frame. The current image is classified into face, no face and uncertain regions by projecting the previous face partition. Therefore, the projection yields the first estimate of the face. The face core component is grown up to obtain the final face segmentation using the previous face extraction process. The technique has been successfully assessed using several test sequences in raw format (MPEG-4 database) as well as in MPEG-1 format (from the MPEG-7 database).     

Towards a system for automatic facial feature detection

A model-based methodology is proposed to detect facial features from a front-view ID-type picture. The system is composed of three modules: context (i.e. face location), eye, and mouth. The context module is a low resolution module which defines a face template in terms of intensity valley regions. The valley regions are detected using morphological filtering and 8-connected blob coloring. The objective is to generate a list of hypothesized face locations ranked by face likelihood. The detailed analysis is left for the high resolution eye and mouth modules. The aim for both is to confirm as well as refine the locations and shapes of their respective features of interest. The detection is done via a two-step modelling approach based on the Hough transform and the deformable template technique. The results show that facial features can be located very quickly with Adequate or better fit in over 80% of the images with the proposed system.     

基于多尺度和多方向特征的人脸超分辨率算法

提出一个基于学习的人脸图像超分辨率算法．该算法采用可操纵金字塔学习人脸图像中的低层次局部特征的空间分布，并结合塔状的父结构和局部最优匹配算法来预测最佳先验模型；然后将先验模型结合到贝叶斯最大后验概率框架中；最后使用最速下降法求出最优的高分辨率人脸图像．实验结果表明，该算法生成的高分辨率人脸图像具有较好的视觉效果．     

Face recognition using the embedded HMM with second-order block-specific observations

The paper is concerned with face recognition using the embedded hidden Markov model (EHMM) with second-order block-specific observations. The proposed method partitions a face image into a 2-D lattice type, composed of many blocks. Each block is represented by the second-order block-specific observation that consists of a combination of first- and second-order feature vectors. The first-order (or second-order) feature vector is obtained by projecting the original (or residual) block image onto the first (or second) basis vector that is obtained block-specifically by applying the PCA to a set of original (or residual) block images. A sequence of feature vectors obtained from the top-to-bottom and the left-to-right scanned blocks are used as an observation sequence to train EHMM. The EHMM models the face image in a hierarchical manner as follows. Several super states are used to model the vertical facial features such as the forehead, eyes, nose, mouth, and chin, and several states in the super state are used to model the localized features in a vertical face feature. Recognition is performed by identifying the person of the model that provides the highest value of observation probability. Experimental results show that the proposed recognition method outperforms many existing methods, such as the second-order eigenface method, the EHMM with DCT observations, and the second-order eigenface method using a confidence factor in terms of average of the normalized modified retrieval rank and false identification rate.     

基于模糊聚类分析的采空区自燃“三带”性质研究

针对采空区内不同区域的煤自燃倾向性存在差异性的问题,采用基于模糊聚类分析的采空区自燃"三带"划分方法对不同测点数据进行了分类,分析了不同λ截距阵下各分类的差异性。分析结果表明:不同的λ值可以将数据划分为不同的类别;在一定范围内,随着λ值的增大,原有的类别可以被划分为更小的子类;当λ值增大时,采空区内距工作面较近区域的测点数据更容易被划分为更小的子类,而距工作面较远区域的测点数据分类较为稳定,则可判定距煤壁较远的采空区区域环境性质较为稳定,而距煤壁较近的采空区区域环境性质差异较大;运用F-统计量计算最优的λ,相应分类与采空区"三带"划分相吻合。     

利用背景先验的显著性检测算法

目的 针对图像的显著区域检测问题,提出一种利用背景先验知识和多尺度分析的显著性检测算法。方法 首先,将原始图像在不同尺度下分解为超像素。然后,在每种尺度下根据各超像素之间的特征差异估计背景,提取背景区域,获取背景先验知识。根据背景先验计算各超像素的显著性,得到显著图。最后,将不同超像素尺度下的显著图进行融合得到最终显著图。结果 在公开的MASR-1000、ECSSD、SED和SOD数据集上进行实验验证,并和目前流行的算法进行实验对比。本文算法的准确率、召回率、F-Measure以及平均绝对误差均在4个数据集上的平均值分别为0.718 9、0.699 9、0.708 6和0.042 3,均优于当前流行的算法。结论 提出了一种新的显著性检测算法,对原始图像进行多尺度分析,利用背景先验计算视觉显著性。实验结果表明,本文算法能够完整、准确地检测显著性区域,适用于自然图像的显著性目标检测或目标分割应用。     

Specific object retrieval based on salient regions

In this paper, we present an image retrieval technique for specific objects based on salient regions. The salient regions we select are invariant to geometric and photometric variations. Those salient regions are detected based on low level features, and need to be classified into different types before they can be applied on further vision tasks. We first classify the selected regions into four types including blobs, edges and lines, textures, and texture boundaries, by using the correlations with the neigbouring regions. Then, some specific region types are chosen for further object retrieval applications. We observe that regions selected from images of the same object are more similar to each other than regions selected from images of different objects. Correlation is used as the similarity measure between regions selected from different images. Two images are considered to contain the same object, if some regions selected from the first image are highly correlated to some regions selected from the second image. Two data sets are employed for experiment: the first data set contains human face images of a number of different people and is used for testing the retrieval algorithm on distinguishing specific objects of the same category; and the second data set contains images of different objects and is used for testing the retrieval algorithm on distinguishing objects of different categories. The results show that our method is very effective on specific object retrieval.     

新闻视频显著人脸检索系统

该文针对新闻视频设计并实现了一个显著人脸检索系统。首先将新闻视频分割成镜头序列,利用训练好的CascadeAdaboost人脸检测器对每个镜头检测出一定数目的候选人脸,按照一些规则选取可信度高的作为样本,用于提取该镜头内的肤色模型。接着对肤色分割后的区域进行位置、大小分析和模板匹配,以淘汰非人脸区域,确定待跟踪的对象列表。为了做精确的跟踪和识别,系统对每个跟踪对象建立更细致的肤色模型。跟踪过程中每间隔一定帧数重新进行人脸检测,以减少误差积累和探测是否有新人脸出现。最后从每个人脸序列挑选最适合进行人脸识别的图像建立其特征脸空间,结合肤色信息和PCA算法判断其是否为要检索的目标人脸。     

一种快速的自动人眼定位方法

眼睛定位是人脸生物特征识别的重要组成部分。提出了一种快速、高效的人眼定位方法,该方法先用基于Haar-like特征的AdaBoost级联分类器检测出人脸块,粗定位左、右眉眼区域,采用改进的灰度投影法分别将两块区域中的眉毛部分分离掉,最后分别将左右眼区域分割成若干小图像块,联合图像块灰度均值、复杂度和结构居中度等信息精确定位人眼。该方法在中科院人脸库CAS-PEAL的Normal子库上表现优异,取得了约99%的定位正确率。实验结果表明,该方法速度快,定位效果好。