Although great progress has been made in generic object detection by advanced deep learning techniques, detecting small objects from images is still a difficult and challenging problem the field of computer vision due to limited size, less appearance, geometry cues, lack large-scale datasets targets. Improving performance wider significance many real-world applications, such as self-driving cars, unmanned aerial vehicles, robotics. In this article, first-ever survey recent studies...

In this paper, we introduce the concept of self-quotient image (SQI) for robust face recognition under varying lighting conditions. It is based on quotient method (Amnon Shashua et al., 2001) (T. Riklin-Raviv 1999) to achieve invariance. However, SQI has three advantages: (1) it needs only one extraction intrinsic invariant property a while removing extrinsic factor corresponding lighting; (2) no alignment needed; (3) works in shadow regions. The theoretical analysis conditions where...

We present a unified framework for modeling intrinsic properties of face images recognition. It is based on the quotient image (QI) concept, in particular existing works QI, spherical harmonic, ratio and retinex. Under this framework, we generalize these previous into two new algorithms: (1) non-point light (NPL-QI) extends QI to deal with sources by directions using harmonic bases; (2) self-quotient (S-QI) perform illumination subtraction without need alignment no shadow assumption....

Text classification is a fundamental task in Nature language process (NLP) application. Most existing research work relied on either explicate or implicit text representation to settle this kind of problems, while these techniques well for sentence and can not simply apply short because its shortness sparseness feature. Given facts that obtaining the simple word vector feature ignoring important by utilizing traditional multi-size filter Convolution neural network (CNN) during course task,...

In recent years, distantly-supervised relation extraction has achieved a certain success by using deep neural networks. Distant Supervision (DS) can automatically generate large-scale annotated data aligning entity pairs from Knowledge Bases (KB) to sentences. However, these DS-generated datasets inevitably have wrong labels that result in incorrect evaluation scores during testing, which may mislead the researchers. To solve this problem, we build new dataset NYTH, where use as training and...

In this paper, we propose a precise and fast eye center localization method. The method is based on SDM [1] algorithm, which has achieved impressive performance in face alignment area. work, further improve the basic algorithm two aspects: 1) employ coarse to fine strategy enhance robustness initialization; 2) utilize multiple nonlinear features ensure while accuracy. Experimental results BioID database shows that our improved achieves preferable accuracy against state-of-the-art methods at...

This paper presents a novel method of camera parameters calibration for obstacle distance measurement based on monocular vision. In this method, resolving have been decomposed into two linear equation sets, which has significantly reduced computing complexity and improved speed. The ideal pinhole imaging model demonstrated, then the equations derived afterwards. end, experiment shows that proposed is very effective.

Based on the analysis of color histogram for image retrieval, a new descriptor, bit-plane distribution entropy (BPDE), is proposed in this paper. The firstly divided into eight bit-planes and Gray-code introduced to avoid effect changes intensity values bit-planes. Then, an vector constructed by computing first four significant planes which contain most structural information image. Finally, with designing correlation-weighted matrix, Mahalanobis distance adopted measure similarity because...

关联特征提取是红外与可见光图像配准、融合和变化检测等应用中的关键步骤。针对同一场景红外与可见光图像间关联特征难以正确提取的问题，基于最稳定极值区域(MSER)算法，提出了一个仿射不变的关联特征提取方法。该方法主要包括3个步骤：(1)提取红外与可见光图像中的最稳定极值区域；(2)对特征区域进行椭圆拟合；(3)规则化处理，消除形变干扰，输出便于描述和匹配的一致性特征。实验结果证明了该算法在红外与可见光图像关联特征提取中的有效性。

We present a new algorithm for recognizing face images under different lighting conditions. Instead of the commonly used methods modeling 3D shape or finding some illumination invariant, basic idea our is to create synthetic by simulating effects input image each candidate in database, and then checking correlation between resulting synthesized images. With method, only single needed which significantly decreases size database. Theory analysis experimental results show that this can...

Textures is one of the keys in image pattern recognition. To realize computer vision for textures, it necessary that understand textures rationally. The key understanding explaining them by human's signs and approaches. We believe mathematical structure categories similar to rational intelligence. In this paper, we describe common notes term theory. It means are useful combination neural network worth research which will gradually reveal unity way percept rationality.

To solve the problems that difficulties to obtain geographical coordinates and meet high performance requirements of fitting equation while making a three-dimensional locus low-speed moving object, GPS-based Locus algorithm is raised in this paper. The algorithm's mathematical model built on parametric equations which do operation according longitude, latitude altitude collected by GPS receiver. Algorithm has feature automatic correction it makes data, received receiver, divided into several...

In order to guarantee key businesses when bandwidth resource is in shortage, an adaptive and flexible Context-Aware Bandwidth Management Scheme (CABMS) proposed improve network survivability. Nodes firstly query the local context information, use Bayesian Network (BN) determine importance of current business, further utility function allocation. Through establishing dual problem original one, "shadow price" introduced, so that nodes are able adjust requests on their own according price, with...

Users engage in various activities different regions of the city, how to infer their roles through area user’s activity, and possible geographical structure role role, it is case challenging issues mobility field. The necessary context elements are extracted from microblogging, such as user ID, check-ins GPS coordinates, mutual inference model urban functional zones established for movement law. experimental results show that has achieved good performance terms using structure, or use roles.