Eye localization and state analysis for driver fatigue detection

To solve the problem of eye precise localization in head multi-position,an algorithm based on skin color and texture feature is proposed. Face is detected based on skin color and according to the gray difference between eyes and other parts,and eye candidate areas are found with the method of binarization and morphological image processing. Eyes are detected by calculating the texture characteristics values of candidate areas. The method of black fleck fitting elliptic is used to check the factuality of the 2 detected eyes and measure the eye opening extent. The experimental results demonstrate the texture characteristics values of candidate areas are still different when head angle changes and the proposed method can accurately detect eyes. The method is simple and fast,and not influenced by driver head posture.     

基于颜色和纹理特征的花生仁外观品质检测研究

为了提高对花生仁外观缺陷的在线分类准确率及效率。通过对采集完好、破损、霉变的花生仁RGB图像进行均值位移法、灰度处理以及阈值分割等预处理,研究提取了花生仁HSV颜色空间下的H、S、V各分量的一阶矩和二阶矩共6个颜色特征值,再基于灰度共生矩阵法提取能量、熵、对比度、逆差分矩共4个纹理特征值,构建颜色和纹理结合的特征向量,最后分别采用BP神经网络和SVM分类器对花生仁进行分类识别。结果表明：在花生仁的整体识别准确率上,BP神经网络为96.67%,SVM分类器为97.22%,后者优于前者,在识别时间上BP和SVM分别为2.5 s和1.1 s,识别效率上也是SVM更好,综合识别准确率和效率两方面考虑,优先选择SVM分类器模型来对花生仁进行分类识别。     

Automatic detection of common surface defects on oranges using combined lighting transform and image ratio methods

Automatic detection of fruit peel defects by a computer vision system is difficult due to the challenges of acquiring images from the surface of spherical fruit and the visual similarity between the stem-ends and the true defects. In this study, oranges with wind scarring, thrips scarring, scale infestation, dehiscent fruit, anthracnose, copper burn, canker spot and normal surface were researched. A lighting transform method based on a low pass Butterworth filter with a cutoff frequency D₀ = 7 was first developed to convert the non-uniform intensity distribution on spherical oranges into a uniform intensity distribution over the whole fruit surface. However, the stem-ends were easily confused with defective areas. In order to solve this problem, different color components (R, G and B) and their combinations were analyzed. It was found that a ratio method and R and G component combination coupled with a big area and elongated region removal algorithm (BER) could be used to differentiate stem-ends from defects effectively. Finally, a processing and classification algorithm based on a simple thresholding method was proposed. The result with 98.9% overall detection rate for the 720 independent sample images indicated that the proposed algorithm was effective in differentiation of normal and defective oranges. The method, however, could not discriminate between different types of defects.     

Affine invariant shape recognition with particle swarmoptimization algorithm

Focusing on the problem that affine transformation will exist among the contour images due to variation of the viewpoints, a new approach to extract affine invariant features and matching strategy is proposed for shape recognition. First, the centroid distance and azimuth angle of each boundary point are computed. Then, with a prior defined angle interval, all the points in the neighbor region of the sample point are considered to calculate the average distance for eliminating noise. After that, the centroid distance ratios(CDRs) of any two contour points with angle difference of 180° are achieved as the representation of the shape, which would be invariant to affine transformation. Since the angles of contour points changed non linearly among affine related images, the CDRs should be resampled to build corresponding relationship. It could be regarded as an optimization problem of path planning. In our method, a PSO based path planning model is presented to address this problem. The experimental results demonstrate the efficiency of the proposed method in shape recognition with translation, scaling, rotation, distortion and noise interference.     

基于粒子群算法和支持向量机的黄花菜叶部病害识别

使用数字图像处理技术,以黄花菜叶部病害图像为识别对象,基于Lab空间和K-means聚类算法分割病害区域,提取目标区域的颜色特征、方向梯度直方图(histogram of oriented gradient,HOG)特征和形状特征,分别建立单一特征模型和特征融合模型,采用粒子群(particle swarm optimization,PSO)算法通过交叉验证优化支持向量机(support vector machine,SVM)模型的惩罚因子和核参数,建立基于PSO-SVM的多特征融合分类模型识别黄花菜病害。基于SVM的多特征融合分类模型识别率高于单一特征分类模型,识别率可达为81.67%;基于PSO-SVM多特征融合分类模型识别率高达92.39%。基于PSO-SVM的多特征分类模型识别率高,可以及时、便捷、高效地识别黄花菜病害。     

多尺度积角点检测和视觉颜色特征的鲁棒车牌定位算法

为解决生鲜产品配送过程中的质量控制问题,将质量功能展开(QFD)与计划-实施-检查-处理(PDCA)循环相结合,运用QFD获得生鲜产品配送质量控制的关键指标,针对关键控制指标,将PDCA循环模型运用到预防措施和纠偏措施的制定和持续改进中,为生鲜产品配送中的质量控制提供指导,从而保证生鲜产品的质量。     

基于RGB线性组合模型的柑橘果实为害状识别

为了准确识别柑橘果实病虫害,提高柑橘生产信息化水平,本研究提出了最优RGB线性组合颜色模型来进行目标识别。首先在具有代表性的几幅图像中选取为害状区域与正常区域的点若干,统计这些样本点的R、G、B值及其均值,设计为害状区域与正常区域灰度差最大线性规划目标函数并求解,建立识别模型。最后利用识别模型结合阈值分割法对采集的96幅柑橘果实图像进行处理,发现识别正确率、误检率、漏检率分别81.25%,14.58%,4.17%,识别效果良好。与2R-G-B模型,G-B模型和R-B模型相比,本方法为害状区域与正常区域灰度差最大,故识别正确率最高,误检率、漏检率最低。试验结果表明：RGB线性组合模型法可用于柑橘果实病虫害的识别。     

A fuzzy fusion approach to gabor transform optimal channels for facial expression recognition

According to its characteristics of multi-direction and multi-scale, Gabor wavelet is divided into 13 channels. Due to different contributions of different channels to the recognition rate, a fuzzy integral fusion approach for facial expression recognition is proposed based on optimal channels. Firstly, three optimal channels are selected according to three proposed principles. And then Gabor features of facial expressional images through those optimal channels are extracted with the dimensions reduction. Finally, each optimal channel is used as a classifier and three classifiers are fused based on multi-classifier combination with fuzzy integral. With the results on JAFFE database, it is found that the recognition rate of the proposed algorithm is 94.41%, which verifies the effectiveness of the proposed algorithm.     

Study on Computer 3D-Reconstruction Algorithm of the Contour Points Matching Interpolation for Serial Cross Images

A computer 3D-reconstruction algorithm was proposed for serial cross images,which is called contour points matching interpolation algorithm. The intermediate contours betweentwo neighbouring contours are preduced by contour points matching interpolation to form 3D-shapeof the reconstructed object. This algorithm is simple and can be easily implemented. Its advantagesare fast speed and strong adaptability for medical serial cross images 3D-reconstruction. Experi-ments show that the results are satisfactory.     

基于多光谱成像技术的小麦品种快速无损鉴定

为了研究多光谱成像技术对小麦品种快速无损鉴定的可行性,采用VideometerLab 多光谱图像采集设备对5 个小麦品种共500 个样品在405~970 nm波段内的进行多光谱图像信息进行采集,获取其光谱、颜色和形态特征。利用主成分分析对5 个小麦品种进行定性鉴别,同时,基于光谱特征和光谱图像特征分别比较了神经网络、支持向量机和随机森林3 种模型的鉴定效果。结果显示：利用19 个光谱特征值建立的模型中,BPNN识别模型效果最佳,其建模集和预测集的识别率分别为100%和91.25%。融合19 个光谱特征和6 个图像特征所建立的模型中,BPNN识别模型效果最佳,其建模集和预测集的识别率分别达到了100%和98.4%。结果表明,基于BPNN的多光谱特征融合能够有效的提高小麦品种鉴定效率,为小麦品种的快速无损检测提供了一个新途径。     

光强倒数色度空间的彩色人脸光照预处理

为了解决彩色人脸识别中色彩信息易受光照影响的问题,提出一种基于光强倒数色度空间(IICS)的彩色人脸图像预处理方法。本方法首先将图像均匀地分割成子块;将每个图像块变换为IICS空间中的一个二维数据集,并根据数据集的线性分布特性估计图像块的光照颜色;然后对全部图像块的光照估计进行颜色直方图统计,根据直方图对分块估计的结果进行合并;最后,利用估计得到的光照和对角模型将图像光照校正到标准白光下,用于人脸识别。在AR和FERET人脸库上的实验表明,通过引入本光照预处理,有效增强了彩色人脸识别方法对光照变化的鲁棒性,提高了识别精度。     

Human Face Location Based on Motion Information in a Complex Background

This paper proposes an improvement for the traditional Optical Algorithm, and presents a new way to image segmentation in a complex background. In addition, combined with the neural network, the system can locate the possible human faces successfully by means of two-step location model. In our system, the searching and locating of the human face is the most important stage. According to this, the authors adopt the two-step way to run, firstly they take up the segmentation of the candidate human face areas and then the accurate face locating based on the neural network is used. This algorithm is fast and robust. Experimental results with real scene images are given out there, and all these prove that two-step method gains many advantages in the course of human face location with motion information, such as real-time, robustness and practicality. In addition, the proposed system is also the fundamental and important part of the perfect human face recognition system.     

Application of Gobor wavelet and SLLE in face recognition

In order to improve the recognition rate of face recognition algorithm, a new algorithm of face recognition is proposed based on Gabor wavelet transform and Supervised Locally Linear Embedding (SLLE). Gabor wavelet is introduced as a method to extract Gabor magnitude features by convolving the normalized face image with multi scale and multi orientation Gabor filters. In the feature extraction module, the dimension of Gabor features is reduced by SLLE. A minimum distance classifier is trained for classification. With the test of the ORL and YALE face database, it is found that 3.5 %~37.8% increase in recognition rate can be achieved compared with other algorithms.     

Feature level fusion of fingerprint and finger vein biometrics based on dynamic weighting

To study the fusion at feature extraction level for fingerprint and finger vein biometrics, a dynamic weighting matching algorithm based on predictive quality evaluation of interest features is proposed. The proposed approach is based on the fusion of the two traits by extracting independent feature point-sets from the two modalities, and making the two point-sets compatible for concatenation. According to the results of features evaluation, dynamic weighting strategy is introduction for the fusion biometrics. The weight of excellent features in fusion is improved, aiming to weaken the influence of low quality and false features so that better effects of fusion can be achieved. Experimental results based on FVC2000 and self-constructed databases of finger vein show that our scheme achieves 98.9% recognition rate, compared with fingerprint recognition and finger vein recognition increased by 6.6% and 9.6% respectively, compared with fusion recognition at matching level increased by 5.4%.     

Single sample face recognition based on B-spline and image gradient

A single sample face recognition algorithm based on B-spline and image gradient is proposed. Image gradient method for face recognition has advantage of illumination invariant. But the recognition rate will be greatly decreased when the image contains noise which will seriously influence gradient information. Traditional methods to reduce noise smooth image at the same time and image gradient reorganization rate will be reduced. As the B-spline filter has the feature which can adjust the order, B-spline filter with different orders can be selected according to the image noise value to minimize noise while preserve image gradient information. Experiments prove that using B-spline and image gradient algorithm can achieve a better recognition rate than traditional filtering method on single sample face recognition problem.     

A face recognition method based on modular PCA and SVD

In order to solve the problem that modular PCA method is sensitive to translation, rotation and other geometric transform, a face recognition method based on modular PCA and singular value decomposition (SVD) is proposed. The PCA features of sub image and SVD features are extracted respectively. The distance measure that fuses information of modular PCA and SVD is obtained. Minimum distance classifier is used to face recognition. Experimental results on ORL human face database show that the proposed method can obtain higher recognition rate.     

Corner detection based on evolution difference of plannar curves in B spline scale space

In order to improve image corner detection precision and efficiency, a novel corner detection algorithm based on B spline scale space evolution difference is proposed. The norm of DoB is defined as corner response function to evaluate the multi scale evolution difference. The DoB corner detector confluents the image boundary features with different scales, which can not only strengthen the response of the feature points, but also depress the influence of noise. Among all corner detection algorithms based on B Spline, the proposed algorithm is relatively lower for computation complexity and faster. The comparative experiments demonstrate that the proposed algorithm works well for localization, robustness against noise, as well as invariance to rotation and scalability.     

An improved morphological edge detection algorithm of medical image

Medical images usually contain much noise which affects the edge detection accuracy. Focusing on this problem, based on the edge detection operator in mathematical morphology, an improved edge detection algorithm is presented by combining the features of the multi structure elements and the multi scale edge detection algorithm. The algorithm performs opening and closing operations on the data with the alternative sequence filters and the structure elements. The weighting operation is applied with different weight coefficients for horizontal, vertical and diagonal directions, while the edge detection operator with dilation type is calculated to obtain the improved edge detection algorithm. The steps of the algorithm are described. The algorithm is used to extract the edge of MRI image as well as the image of Lena. The experimental results indicate the algorithm can considerably improve the edge resolution of the traditional morphological edge detection methods and is practical.