基于多纹理特征融合的震后SAR图像倒塌建筑物信息提取

合成孔径雷达（SAR）凭借其全天候观测能力以及SAR图像中丰富的纹理信息,在震后建筑物倒塌评估中发挥了重要作用。针对SAR图像中倒塌建筑物纹理特征多样但利用率较低,且特征信息冗余的问题,提出一种基于主成分分析的SAR图像多纹理特征分类方法。该方法基于灰度直方图、灰度共生矩阵、局部二值模式、Gabor滤波器提取了26种纹理特征信息,构建主成分变量进行多维特征优选与降维融合,通过随机森林分类算法提取建筑物的倒塌信息。以2016年日本熊本地震为例验证了该方法的有效性,结果显示其提取精度高达79.85%,倒塌建筑物的识别效率有所提高,分类结果优于单种纹理特征提取方法及多种纹理特征组合提取法,可用于震后建筑物震害信息的快速提取。     

基于Self-Attention StyleGAN的皮肤癌图像生成与分类

针对以黑色素瘤为代表的皮肤癌分类任务存在数据集各类样本数量、权重不均衡,且现有的对抗生成网络生成的皮肤癌样本图像质量较差导致临床诊断时难以分辨等问题,提出了一种基于自注意力的样式生成对抗网络（Self-Attention StyleGAN）与SE-ResNeXt-50相结合的皮肤癌图像样本生成与分类框架。该框架在样式生成对抗网络（StyleGAN）的基础上引入了自注意力机制,对生成器的样式控制和噪声输入结构进行了重新设计,并重构了鉴别器对图像生成器进行了调整,从而有效地合成高质量的皮肤癌病变图像。使用SE-ResNeXt-50来对皮肤癌样本图像进行分类,更好地提取样本图像不同层次特征图的信息,从而提高了平衡多类精度（BMA）。实验结果表明,该模型在ISIC2019皮肤癌数据集上生成的样本图像质量较高,且分类BMA达到94.71%。该方法提高了皮肤癌病变图像分类的准确性,帮助皮肤科医生对不同类型的皮肤癌病变进行判断和诊断,并对不同阶段和难以区分的皮肤癌病变进行分析。     

基于近邻三值模式和协作表示的三维掌纹识别

针对二维掌纹图像存在易伪造、抗噪能力差的问题，提出一种基于近邻三值模式（NTP）和协作表示的三维掌纹识别方法。首先，利用形状指数把三维掌纹的表面几何信息映射成二维数据，以弥补常用均值或高斯曲率映射无法精确描述三维掌纹特征的缺陷；其次，对形状指数图作分块处理，利用近邻三值模式提取分块形状指数图的纹理特征；最后，利用协作表示的方法进行特征分类。在三维掌纹库上和经典算法进行的对比实验中，该方法的识别率为99.52%，识别时长为0.6738 s，优于其他算法；在识别率方面，与经典的局部二值模式（LBP）、局部三值模式（LLTP）、CompCode、均值曲率图（MCI）法相比分别提高了7.77%、6.02%、5.12%和3.97%；在识别时间方面，与Homotopy、对偶增广拉格朗日法（DALM）、SpaRSA方法相比分别降低了6.7 s、15.9 s和61 s。实验结果表明，所提算法具有良好的特征提取和分类能力，能够有效地提高识别精度并减少识别时间。     

A retrieval system to analyse dermatological lesions using feature ortho-normalisation

An information retrieval system is proposed as an assistance tool for diagnosing the skin lesion using Content-Based Image Retrieval approach. Efficiency of the retrieval system is deliberated in terms of the most relevant retrieval of images from database. The proposed diagnostic assistive model retrieves the skin lesion images and its disease category, case history, symptoms and treatment plan. This retrieval process is made from a dermatology database by the way of visual features in the input image such as shape, texture and colour. The author’s proposed principal component analysis (PCA) feature projection technique is to discriminate the features by projecting them onto a feature subspace. While projecting the features onto a feature subspace features are normalised orthogonally. So the proposed methodology is used to improve the classification by the way of discriminate the features, in-turn it focus the retrieval of comprehensive reference sources, so that the diagnosis accuracy of the dermatologists are also improved. Receiver-operating characteristic curve is used to analyse the proposed computer-aided diagnosis (CAD) method, while analysis we attained high contribution to detect the skin lesions. Totally 1450 images are experimented and the system produced the 99.09% specificity, 96.69% sensitivity and 98.3% accuracy. When compared with other works this system of assessment shows high retrieval and diagnosis concert.     

基于Inception深度残差网络的皮肤黑色素癌图像分类算法

由于皮肤黑色素癌图像存在类内差异大、样本数据集小等特点, 采用深度残差网络可以有效解决训练过程中过拟合问题, 提高识别准确率. 但是深度残差网络模型的训练参数多, 时间复杂度高. 为了提高训练效率, 提高识别准确率, 首先从理论上分析了深度残差网络模型的结构, 通过修改网络结构, 利用Inception结构代替残差网络中的卷积层、池化层, 减少模型的训练参数数量, 降低时间复杂度. 在此基础上, 提出了基于Inception深度残差网络皮肤黑色素癌分类识别算法(Inception Deep Residual Network, IDRN), 用Inception结构代替残差网络中的卷积池化层, 用SeLU激活函数代替传统的ReLU函数. 之后, 在公开的黑色素癌皮肤镜图像ISIC2017数据集上进行实验验证. 理论和实验表明, 与传统的卷积神经网络ResNet50相比, 本文提出的新的分类算法降低了时间复杂度, 提高了识别准确率.     

Toward a combined tool to assist dermatologists in melanoma detection from dermoscopic images of pigmented skin lesions

In this paper we propose a machine learning approach to classify melanocytic lesions as malignant or benign, using dermoscopic images. The lesion features used in the classification framework are inspired on border, texture, color and structures used in popular dermoscopy algorithms performed by clinicians by visual inspection. The main weakness of dermoscopy algorithms is the selection of a set of weights and thresholds, that appear not to be robust or independent of population. The use of machine learning techniques allows to overcome this issue. The proposed method is designed and tested on an image database composed of 655 images of melanocytic lesions: 544 benign lesions and 111 malignant melanoma. After an image pre-processing stage that includes hair removal filtering, each image is automatically segmented using well known image segmentation algorithms. Then, each lesion is characterized by a feature vector that contains shape, color and texture information, as well as local and global parameters. The detection of particular dermoscopic patterns associated with melanoma is also addressed, and its inclusion in the classification framework is discussed. The learning and classification stage is performed using AdaBoost with C4.5 decision trees. For the automatically segmented database, classification delivered a specificity of 77% for a sensitivity of 90%. The same classification procedure applied to images manually segmented by an experienced dermatologist yielded a specificity of 85% for a sensitivity of 90%.     

融合颜色特征与LBP的图像检索

随着图像数据量的不断丰富和人们需求的不断提高,基于内容的图像检索技术成为一个重要研究课题。研究了基于颜色相关图与LBP算法的图像检索算法,利用颜色相关图与LBP算法分别提取图像的颜色特征与纹理特征,实现了图像的检索,实验结果说明了算法的有效性。     

Color based skin classification

Skin detection is used in applications ranging from face detection, tracking body parts and hand gesture analysis, to retrieval and blocking objectionable content. In this paper, we investigate and evaluate (1) the effect of color space transformation on skin detection performance and finding the appropriate color space for skin detection, (2) the role of the illuminance component of a color space, (3) the appropriate pixel based skin color modeling technique and finally, (4) the effect of color constancy algorithms on color based skin classification. The comprehensive color space and skin color modeling evaluation will help in the selection of the best combinations for skin detection. Nine skin modeling approaches (AdaBoost, Bayesian network, J48, Multilayer Perceptron, Naive Bayesian, Random Forest, RBF network, SVM and the histogram approach of Jones and Rehg (2002)) in six color spaces (IHLS, HSI, RGB, normalized RGB, YCbCr and CIELAB) with the presence or absence of the illuminance component are compared and evaluated. Moreover, the impact of five color constancy algorithms on skin detection is reported. Results on a database of 8991 images with manually annotated pixel-level ground truth show that (1) the cylindrical color spaces outperform other color spaces, (2) the absence of the illuminance component decreases performance, (3) the selection of an appropriate skin color modeling approach is important and that the tree based classifiers (Random forest, J48) are well suited to pixel based skin detection. As a best combination, the Random Forest combined with the cylindrical color spaces, while keeping the illuminance component outperforms other combinations, and (4) the usage of color constancy algorithms can improve skin detection performance.     

Breast density classification to reduce false positives in CADe systems

This paper describes a novel weighted voting tree classification scheme for breast density classification. Breast parenchymal density is an important risk factor in breast cancer. Moreover, it is known that mammogram interpretation is more difficult when dense tissue is involved. Therefore, automated breast density classification may aid in breast lesion detection and analysis. Several classification methods have been compared and a novel hierarchical classification procedure of combined classifiers with linear discriminant analysis (LDA) is proposed as the best solution to classify the mammograms into the four BIRADS tissue classes. The classification scheme is based on 298 texture features. Statistical analysis to test the normality and homoscedasticity of the data was carried out for feature selection. Thus, only features that are influenced by the tissue type were considered. The novel classification techniques have been incorporated into a CADe system to drive the detection algorithms and tested with 1459 images. The results obtained on the 322 screen-film mammograms (SFM) of the mini-MIAS dataset show that 99.75% of samples were correctly classified. On the 1137 full-field digital mammograms (FFDM) dataset results show 91.58% agreement. The results of the lesion detection algorithms were obtained from modules integrated within the CADe system developed by the authors and show that using breast tissue classification prior to lesion detection leads to an improvement of the detection results. The tools enhance the detectability of lesions and they are able to distinguish their local attenuation without local tissue density constraints.     

单极化合成孔径雷达图像颜色特征编码与分类

针对目前单极化合成孔径雷达（SAR）伪彩色编码方法存在的细节信息和可视性不强的问题,提出一种颜色特征编码方法。该颜色特征编码方法首先对单极化SAR图像提取纹理特征;然后将每一个特征量化到0到255;其次对每一个灰度级赋予一个RGB颜色,编码成颜色特征图;最后对随机森林计算得到的特征重要性进行排序,每3维特征对应为R、G、B通道生成伪彩图。基于该颜色特征编码方法,提出一种新的分类方法。该分类方法首先根据目视效果选择可分性最好的伪彩图;然后采用统计区域合并（SRM）分割算法对其分割;其次将所有RGB伪彩图作为分类的特征,以随机森林为分类器进行分类,得到初步的结果;最后对初步的结果进行相对多数投票,得到最终的分类结果。方法验证采用两组TerraSAR-X单极化SAR数据,与基于HIS的颜色编码方法对比,该颜色特征编码方法生成的伪彩图信息熵得到了很大提升,且两组数据每类地物的分类精度都大幅度提高,因此证明了所提算法保留了更多的细节信息,获取更多的颜色信息,更利于可视化和地物分类,从而表明提出的颜色特征编码方法是可行的。     

对立色LBP模型的目标跟踪

目标表示方法对跟踪方法的鲁棒性有着重要影响。将对立色局部二值模式(OCLBP)纹理算子作为研究对象引入目标表示。通过分析不同颜色通道之间的相关性和OCLBP的10种纹理模式的表征能力,选择目标候选区域中具有OCLBP的7种主要模式的关键点的纹理直方图作为目标模型。最后将该目标表示方法嵌入到MeanShift框架中,进行目标跟踪。实验结果表明,提出的基于OCLBP主要模式的目标表示方法显著提高了Mean Shift目标跟踪方法的性能。