Kernel-aligned multi-view canonical correlation analysis for image recognition

Existing kernel-based correlation analysis methods mainly adopt a single kernel in each view. However, only a single kernel is usually insufficient to characterize nonlinear distribution information of a view. To solve the problem, we transform each original feature vector into a 2-dimensional feature matrix by means of kernel alignment, and then propose a novel kernel-aligned multi-view canonical correlation analysis (KAMCCA) method on the basis of the feature matrices. Our proposed method can simultaneously employ multiple kernels to better capture the nonlinear distribution information of each view, so that correlation features learned by KAMCCA can have well discriminating power in real-world image recognition. Extensive experiments are designed on five real-world image datasets, including NIR face images, thermal face images, visible face images, handwritten digit images, and object images. Promising experimental results on the datasets have manifested the effectiveness of our proposed method.     

图像中近圆形对象的一种识别方法

介绍了一种利用多特征联合概率的红外图像中近圆形对象检测方法,通过计算图像中诸多小段曲线的弯拱方向来获得图像中各处存在圆的概率,并用梯度强度对这些概率进行加权,再结合对象内部为空洞的概率,共同形成对图像中近圆形对象所在位置的一个判决。经与经典的Hough变换法相比较,表现出一定的优越性。对一组不同特征的红外图像进行测试实验,均能检测出近圆形对象,表现出一定的适应性,对某些应用场景中的目标检测与识别有辅助价值。     

改进粒子群优化算法的高光谱波段选择

高光谱图像具有数百个连续、狭窄的光谱带,光谱范围跨越可见光到红外光,可提供地物的精细光谱属性,对于地物材质和属性的识别分类具有重要应用价值。针对感兴趣目标选择有限的光谱波段进行传输和处理,对于提升高光谱数据处理时效性、以及设计面向特定应用的实用化光谱仪都具有重要意义。而如何结合目标特征选择最优波段成为在提升处理效率的同时保证目标识别或分类精度的必然要求。因此如何从数以百计维度的高光谱图像中选择出具有较好分类识别能力的波段子集是急需解决的问题。提出基于改进粒子群优化算法的高光谱波段选择方法,该方法区别于传统的粒子群优化算法,引入 “概率突跳特性”,并设定新解的淘汰机制,将“停滞”的新解进行淘汰,提高了算法的全局寻优性能。然后基于目标光谱特征采用了最优波段选择的优化目标函数,通过改进的粒子群优化算法求解目标函数,并将选定的波段子集反馈到支持向量机（SVM）中执行分类应用。采用两个标准的高光谱数据集（Indian Pines, Salinas）对选择出的波段子集进行分类测试,结果表明该方法相较于现有方法具有较高的分类精度,在几种方法中,传统的粒子群算法筛选出的波段效果最差;该算法筛选出的波段的分类精度最好,两个数据集的分类精度分别可以达到98.141 4%和99.084 8%。     

基于传像束的小视场双目结构光三维形貌恢复

提出基于传像束传像和双目结构光相结合的三维测量系统,实现对无法直接成像场景,如爆轰、冲击等环境下的小视场物体三维面形测量.测量时,结构光直接投影到待测物体表面,变形条纹图像首先被成像到传像束前端面,再被传递到传像束后端面,并最终被成像到相机像平面,再由三频相位展开技术计算出对应相位分布后,即可根据双目匹配计算出待测物体...     

Occlusion removal technique for improved recognition of partially occluded 3D objects in computational integral imaging

In this paper, we propose an occlusion removal technique for improved recognition of 3D objects that are partially occluded in computational integral imaging (CII). In the reconstruction process of a 3D object which is partially occluded by other objects, occlusion degrades the resolution of reconstructed 3D images and thus this affects negatively the recognition of a 3D object in CII. To overcome this problem, we introduce a method to eliminate occluding objects in elemental image array (EIA) and the proposed method is applied to 3D object recognition by use of CII. To our best knowledge, this is the first time to remove occlusion in CII. In our method, we apply the elemental image to sub-image (ES) transform to EIA obtained by a pickup process and those sub-images are employed for occlusion removal. After the transformation, we correlate those sub-images with a reference sub-image to locate occluding objects and then we eliminate the objects. The inverse ES transform provides a modified EIA. Actually, the modified EIA is considered to be an EIA without the object that occludes the object to be reconstructed. This can provide a substantial gain in terms of the image quality of 3D objects and in terms of recognition performance. To verify the usefulness of the proposed technique, some experimental results are carried out and the results are presented.     

Stokes参量数字全息法实现相位物体成像

为实现对相位物体的无损检测和成像,克服数字同轴全息相位物体成像技术在消除零级像和孪生像的干扰时存在的系列问题,提出一种基于Stokes参量的新的数字同轴全息技术。该方法区别于传统的利用干涉光场来记录原始像项的数字全息方法,通过测量物参光合成光束的Stokes参量来分别得到这两束光的振幅和相位差,从而准确、唯一地获得原始像项;再利用数字再现即可重构物光的振幅和相位信息。实验中对弱吸收的相位样品进行了测量,得到样品清晰的振幅和相位分布。结果表明,采用该方法对相位物体进行数字全息再现,可以克服传统同轴全息图中零级像和共轭像对相位物体信息的严重干扰,对于提取相位物体的振幅和相位信息是可行和有效的。     

Optical technique for classification, recognition and identification of obscured objects

The capability to classify, recognize and to identify objects from spatially low resolution images has high significance in security related applications especially in a case that recognition of camouflaged object is required.In this paper we present a novel approach in which the scenery containing obscured objects which we wish to classify, recognize or identify is illuminated by spatially coherent beam (e.g. laser) and therefore secondary speckles pattern is reflected from the objects. By special image processing algorithm developed for this research and which is basically based upon temporal tracking of the random speckle pattern one may extract the temporal signature of the object. And right after, to use it for its classification (e.g. its separation from the other objects in the scenery), its recognition and identification even in a case that the imager provides poor spatial resolution that by itself does not allow doing the specified detection related operations.     

Enhanced computational integral imaging system for partially occluded 3D objects using occlusion removal technique and recursive PCA reconstruction

In this paper, we propose an enhanced computational integral imaging system by both eliminating the occlusion in the elemental images recorded from the partially occluded 3D object and recovering the entire elemental images of the 3D object. In the proposed system, we first obtain the elemental images for partially occluded object using computational integral imaging system and it is transformed to sub-images. Then we eliminate the occlusion within the sub-images by use of an occlusion removal technique. To compensate the removed part from occlusion-removed sub-images, we use a recursive application of PCA reconstruction and error compensation. Finally, we generate the entire elemental images without a loss from the newly reconstructed sub-images and perform the process of object recognition. To show the usefulness of the proposed system, we carry out the computational experiments for face recognition and its results are presented. Our experimental results show that the proposed system might improve the recognition performance dramatically.     

Synthesis of three-dimensional phase images of nanoobjects: Numerical simulation

A way to synthesize three-dimensional phase-diffraction images of nanoobjects is proposed based on the numerical method of backward propagation of a field from the best focusing plane. The obtained phase images are used for solving the applied problem of object defect finding by methods of pattern recognition. The probability of defect detection by phase images is shown to be higher than for amplitude images.     

基于微透镜阵列的实时三维物体识别

提出一种基于微透镜阵列多视角成像特点,将三维物体的深度信息转化为二维透射像阵列的角度信息,利用光学二维图像识别技术,实现对三维物体识别的方法.对识别过程进行了理论分析和计算,用匹配滤波的方法实现了对三维物体骰子的实时识别.实验结果表明,本方法的相关识别能力较高,并且具有很强的灵活性,对于有微小旋转、微小平移的三维物体也可进行识别.     

利用偏振技术识别人造目标

提出了一种利用目标的偏振信息识别人造目标的新型方法。利用自制的多波段偏振CCD地面实验装置获取目标的偏振图像,并提取其中的偏振信息。由于人造目标和自然目标的偏振特性上有较大差别,因而根据这些信息,通过较常规的图像处理手段,即可很好地识别出图像中的人造目标。实验证明,该方法识别自然背景下的人造目标是相当有效的。     

结合区域提取和改进卷积神经网络的水下小目标检测

针对水下小目标信息量有限而难以提取有效特征导致的检测性能不佳问题,提出了一种结合区域提取和融合Hu矩特征的改进卷积神经网络水下小目标检测方法。该方法包含区域提取和分类两个步骤。首先以马尔可夫随机场分割算法为基础进行区域提取,对潜在目标定位的同时降低伪目标对后续分类的干扰;然后提取潜在目标区域的Hu矩特征并融入卷积神经网络,形成一种形状特征表征能力更强的改进卷积神经网络用于分类。声呐实测数据处理结果表明,该方法可以有效提升对水下小目标的发现概率和正确报警率,与其他目标检测方法相比,该方法具有更好的检测性能和泛化性。     

Producing virtual face images for single sample face recognition

For single sample face recognition, there are limited training samples, so the traditional face recognition methods are not applicable to this problem. In this paper we propose to combine two methods to produce virtual face images for single sample face recognition. We firstly use a symmetry transform to produce symmetrical face images. We secondly use the linear combination of two samples to generate virtual samples. As a result, we convert the special single sample problem into a non-single sample problem. We then use the 2DPCA method to extract features from the samples and use the nearest neighbor classifier to perform classification. Experimental results show that the proposed method can effectively improve the recognition rate of single sample face recognition.     

Self-Adaptive Image Thresholding within Nonextensive Entropy and the Variance of the Gray-Level Distribution

In order to automatically recognize different kinds of objects from their backgrounds, a self-adaptive segmentation algorithm that can effectively extract the targets from various surroundings is of great importance. Image thresholding is widely adopted in this field because of its simplicity and high efficiency. The entropy-based and variance-based algorithms are two main kinds of image thresholding methods, and have been independently developed for different kinds of images over the years. In this paper, their advantages are combined and a new algorithm is proposed to deal with a more general scope of images, including the long-range correlations among the pixels that can be determined by a nonextensive parameter. In comparison with the other famous entropy-based and variance-based image thresholding algorithms, the new algorithm performs better in terms of correctness and robustness, as quantitatively demonstrated by four quality indices, ME, RAE, MHD, and PSNR. Furthermore, the whole process of the new algorithm has potential application in self-adaptive object recognition.     

Co-Training for Visual Object Recognition Based on Self-Supervised Models Using a Cross-Entropy Regularization

Automatic recognition of visual objects using a deep learning approach has been successfully applied to multiple areas. However, deep learning techniques require a large amount of labeled data, which is usually expensive to obtain. An alternative is to use semi-supervised models, such as co-training, where multiple complementary views are combined using a small amount of labeled data. A simple way to associate views to visual objects is through the application of a degree of rotation or a type of filter. In this work, we propose a co-training model for visual object recognition using deep neural networks by adding layers of self-supervised neural networks as intermediate inputs to the views, where the views are diversified through the cross-entropy regularization of their outputs. Since the model merges the concepts of co-training and self-supervised learning by considering the differentiation of outputs, we called it Differential Self-Supervised Co-Training (DSSCo-Training). This paper presents some experiments using the DSSCo-Training model to well-known image datasets such as MNIST, CIFAR-100, and SVHN. The results indicate that the proposed model is competitive with the state-of-art models and shows an average relative improvement of 5% in accuracy for several datasets, despite its greater simplicity with respect to more recent approaches.     

基于可变形部件模型HOG特征的人形目标检测

使用单幅图像进行特定目标的检测是机器视觉领域的重要任务之一。利用机器学习的方法,使用LSVM分类器进行人形目标的检测。该方法提取图像的HOG（梯度方向直方图）特征和其对应的可变形部件来描述目标的外形特征,能够较好地解决目标由于运动而产生外形变化的问题。对常见公共区域场景进行数据采集并随机抽取了200张图像,使用所述方法对其中共1 100个人形目标进行检测,正确率识别率为78.3%。结果表明该方法具有一定的可行性和稳定性,能够较好检测出单幅图像中的人形目标并加以标注。但对于某种程度有所遮蔽的人形目标则会产生漏检的现象。     

Object recognition by use of polarimetric phase-shifting digital holography

Pattern recognition by use of polarimetric phase-shifting digital holography is presented. Using holography, the amplitude distribution and phase difference distribution between two orthogonal polarizations of three-dimensional (3D) or two-dimensional phase objects are obtained. This information contains both complex amplitude and polarimetric characteristics of the object, and it can be used for improving the discrimination capability of object recognition. Experimental results are presented to demonstrate the idea. To the best of our knowledge, this is the first report on 3D polarimetric recognition of objects using digital holography.     

Recognition of a light line pattern by Hu moments for 3-D reconstruction of a rotated object

A simple technique of a light line projection for 3-D shape detection of rotated objects is presented. In this technique, an object is rotated around its symmetrical axis four times at an angle by using an electromechanical device and scanned by a light line. Four views of the object surface are extracted from each one of these rotations by processing a set of light line images. These views are connected using rotation angle and origin coordinates to obtain the complete 3-D shape. Angle and origin are calculated by recognition of a light line pattern. Light line pattern is recognized by Hu moments. In this manner, measurement errors on setup are avoided. It is an advantage over common methods, where these two parameters are measured directly on the setup to obtain the 3-D shape. Local profilometric method is based on the perturbation that the light line suffers when it is projected on the object surface. This perturbation is observed on an image plane due to the different direction between light line projector and viewer. These perturbations are measured by using Gaussian functions. In this technique the light line images are processed in very fast form. The technique and processing time are presented in detail. This technique is tested with objects, which have little information and its experimental results are also presented.     

Modular Fisher Discriminant Sparse Representation for robust face recognition

This paper proposes a novel framework for robust face recognition based on sparse representation and discrimination ranking. This method consists of three stages. The first stage partitions each training sample into some overlapped modules and then computes each module's Fisher ratio, respectively. The second stage selects modules which have higher Fisher ratios to comprise a template to filter training and test images. The dictionary is constructed by the filtered training images. The third stage computes the sparse representation of filtered test sample on the dictionary to perform identification. The advantages of the proposed method are listed as follows: the first stage can preserve the local structure. The second stage removes the modules that have little contribution for classification. Then the method uses the retaining modules to classify the test sample by SRC which makes the method robust. Compared with the related methods, experimental results on benchmark face databases verify the advancement of the proposed method. The proposed method not only has a high accuracy but also can be clearly interpreted.