一种基于彩色特征点的对象查询方法

图像检索系统中对象的查询,主要采用分割技术来实现。虽然一些研究人员提出过利用光学特征的方法,但是只应用于灰度图像,不能对大量的彩色图像进行有效的查询。为此本文提出了一种基于彩色特征点的对象查询新方法。首先,利用Harris彩色点提取器从彩色图像中提取出一组能表征图像特征的彩色点;其次,利用特征点的梯度、角度不变性构造检索策略来实现对象的查询。实验表明,该方法对于对象的光照变化和几何变化有较强的鲁棒性。

面向驾驶场景的多尺度特征融合目标检测方法

针对驾驶场景中目标检测卷积神经网络模型检测精度较低的问题,提出一种基于改进RefineDet网络结构的多尺度特征融合目标检测方法。在RefineDet网络结构中嵌入LFIP(Light-weightFeaturizedImagePyramid,轻量级特征化的图像金字塔)网络,将LFIP网络生成的多尺度特征图与RefineDet中的ARM(AnchorRefinement Module,锚点框修正模块)输出的主特征图相融合,提升特征层中锚点框初步分类和回归的输出效果,为ODM(ObjectDetectionModule,目标检测模块)模块提供修正的锚点框以便于进一步回归和多类别预测;在RefineDet网络结构中的ODM之后嵌入多分支结构RFB(ReceptiveFieldBlock,感受野模块),在检测任务中获得不同尺度的感受野以改善主干网络中提取的特征。将模型中的激活函数替换为带有可学习参数的非线性激活函数PReLU(Parametric RectifiedLinearUnit,参数化修正线性单元),加快网络模型的收敛速度;将RefineDet的边界框回归损失函数替换为排斥力损失函数RepulsionLoss,使目标检测中的某预测框更靠近其对应的目标框,并使该预测框远离附近的目标框及预测框,可以提升遮挡情况下目标检测的精度;构建驾驶视觉下的目标检测数据集,共计48260张,其中38608张作为训练集,9652张作为测试集,并在主流的GPU硬件平台进行验证。该方法的mAP为85.59%,优于RefineDet及其他改进算法;FPS为41.7 frame/s,满足驾驶场景目标检测的应用要求。实验结果表明,该方法在检测速度略微下降的情况,能够较好地提升驾驶视觉下的目标检测的精确度,并能够一定程度上解决驾驶视觉下的遮挡目标检测和小目标检测的问题。

Coordinated optimization setting of reagent dosages in roughing-scavenging process of antimony flotation

Considering the influence of reagent adjustment in different flotation bank on the final production index and the difficulty of establishing an effective mathematical model,a coordinated optimization method for dosage reagent based on key characteristics variation tendency and case-based reasoning is proposed.On the basis of the expert reagent regulation method in antimony flotation process,the reagent dosage pre-setting model of the roughing–scavenging bank is constructed based on case-based reasoning.Then,the sensitivity index is used to calculate the key features of reagent dosage.The reagent dosage compensation model is constructed based on the variation tendency of the key features in the roughing and scavenging process.At last,the prediction model is used to finish the classification and discriminant analysis.The simulation results and industrial experiment in antimony flotation process show that the proposed method reduces fluctuation of the tailings indicators and the cost of reagent dosage.It can lay a foundation for optimizing the whole process of flotation.

CT image morphology features of pulmonary sclerosing hemangiomas

Objective: The aim of the study was to analyze the CT morphology features of pulmonary sclerosing hemangiomas (PSHs) and improve the diagnosis ability of this disease.Methods: The 18 cases of pulmonary sclerosing hemangioma (PSH) confirmed by operation and histopathology from August 2002 to May 2009 were collected,including 17 females and 2 males,aged from 19 to 60 years old,with an average age of 43 years.All the cases underwent plain CT scan,among them,16 cases received enhanced CT scan.Results: The 18 cases had isolated mass.Mean long-axis diameter of these lesions was (2.7 ± 1.3) cm (range,1.9–4.2 cm).Of all cases,5 cases (27.8%) were round in shape,9 cases (50%) were oval,4 cases (22.2%) were lobulated,and 14 cases (77.8%) were smooth margin.The air meniscus sign was in 2 cases (11.1%),and the halo sign in 3 cases (16.7%).Two cases (11.1%) contained small nodular calcification,the remaining 16 cases (70%) were homogeneous density,the CT density of the masses ranged from 24–47 HU,and the mean value was 35 HU.Sixteen cases received enhanced scan,the welt vessel sign was in 8 cases (44.4%),1 case showed less enhancement,5 cases showed marked homogeneous enhancement and 10 cases showed intense and patchy heterogeneous enhanced.The CT density of the enhancing masses ranged from 60–110 HU,the mean value was 35 HU,and the net enhancement value was 14–80 HU,the mean value was 55 HU.Conclusion: PSH should be considered in middle-aged female whose CT found that single round or oval pulmonary nodules,with smooth margin,or associated with the air meniscus sign,the halo sign,or the marked enhancement.

No-Reference Quality Assessment of Enhanced Images

Image enhancement is a popular technique,which is widely used to improve the visual quality of images.While image enhancement has been extensively investigated,the relevant quality assessment of enhanced images remains an open problem,which may hinder further development of enhancement techniques.In this paper,a no-reference quality metric for digitally enhanced images is proposed.Three kinds of features are extracted for characterizing the quality of enhanced images,including non-structural information,sharpness and naturalness.Specifically,a total of 42 perceptual features are extracted and used to train a support vector regression(SVR) model.Finally,the trained SVR model is used for predicting the quality of enhanced images.The performance of the proposed method is evaluated on several enhancement-related databases,including a new enhanced image database built by the authors.The experimental results demonstrate the efficiency and advantage of the proposed metric.

Blur Invariant Image Forgery Detection Method Using Local Phase Quantization

With the rapid development of powerful image, editing software makes the forgery of the digital image easy. Researchers proposed methods to cope with image authentication in recent years. We proposed a passive image authentication technique to determine the copy move forgery that copied a part of an image and pasted it on the other region in the same image. First, the method divides the image into overlapping blocks. It uses LPQ （local phase quantization） to label each block. The column average value of labeled blocks constitutes the feature vector for the block. Similarity among the feature vectors gives a clue about the forgery. Local phase quantization has not been used to detect copy move forgery in the literature before. Experimental results show that, the method has higher accuracy ratios and lower false negative values under blurring operation at high levels compared to other methods. Our method can also detect multiple copy move forgery.

Image feature optimization based on nonlinear dimensionality reduction

Image feature optimization is an important means to deal with high-dimensional image data in image semantic understanding and its applications. We formulate image feature optimization as the establishment of a mapping between highand low-dimensional space via a five-tuple model. Nonlinear dimensionality reduction based on manifold learning provides a feasible way for solving such a problem. We propose a novel globular neighborhood based locally linear embedding (GNLLE) algorithm using neighborhood update and an incremental neighbor search scheme, which not only can handle sparse datasets but also has strong anti-noise capability and good topological stability. Given that the distance measure adopted in nonlinear dimensionality reduction is usually based on pairwise similarity calculation, we also present a globular neighborhood and path clustering based locally linear embedding (GNPCLLE) algorithm based on path-based clustering. Due to its full consideration of correlations between image data, GNPCLLE can eliminate the distortion of the overall topological structure within the dataset on the manifold. Experimental results on two image sets show the effectiveness and efficiency of the proposed algorithms.

Temporal and spatial analysis of changes in snow cover in western Sichuan based on MODIS images

We developed a method for analyzing the change in snow cover using MODIS imagery.The method was applied to images of western Sichuan Province,China taken between 2002 and 2008.The model for extracting data on snow cover from MODIS images was created by spectral analysis.The multi-temporal snow layers were used to evaluate the temporal and spatial change in the area under snow cover between 2002 and 2008 using overlay and statistical analysis in ARCGIS.The majority(60.4%) of western Sichuan was rarely covered by snow and only 0.3% was covered by perennial snow in 2002.Snow cover was pri-marily distributed in Garzê and Aba.The area under snow cover was significantly and negatively correlated with the average monthly temperature and rainfall in 2002.The largest area under snow cover was measured in 2006 and the smallest was in 2007.Similarly,the area of snowmelt was the highest in 2006 and lowest in 2007.In general,the elevation of the snow line in-creased throughout the period 2002-2008;however,the elevation decreased in some years.Our results provide an important insight into the distribution of snow in this region,and may be useful for climate modeling and predicting the availability of water resources and the occurrence of floods and droughts.