灰度图像交叉熵阈值法的三维推广

针对现有的灰度图像交叉熵阈值化方法无法有效分割含有混合噪声图像的问题,在图像三维直方图的基础上提出三维交叉熵阈值化算法,同时给出三维交叉熵阈值法的快速递推公式.实验结果表明,三维方法结合了图像中像素的灰度及其局部空间的均值和中值信息,对于含有混合噪声的图像,具有比现有交叉熵阈值化算法更好的分割效果.     

Effectiveness evaluation of iris segmentation by using geodesic active contour (GAC)

A novel iris segmentation technique based on active contour is proposed in this paper. Our approach uses innovative algorithms, including two important ones, pupil segmentation and iris circle calculation. With our algorithms, we are able to find the center position and radius of pupil correctly and segment the iris precisely. The accuracy of our proposed method for ICE dataset is around 92% and also reached high accuracy level of 79% for UBIRIS. Our results demonstrate that the proposed iris segmentation method can perform well with high accuracy and better efficacy for Iris segmentation in images. Through a relatively high-performance algorithm to further cut up the round out the picture of the pupil conversion cutting growth square picture in order to make the judgment for biometric applications.

     

Improving image segmentation based on patch-weighted distance and fuzzy clustering

Image segmentation is the basis of image analysis, object tracking, and other fields. However, image segmentation is still a bottleneck due to the complexity of images. In recent years, fuzzy clustering is one of the most important selections for image segmentation, which can retain information as much as possible. However, fuzzy clustering algorithms are sensitive to image artifacts. In this study, an improved image segmentation algorithm based on patch-weighted distance and fuzzy clustering is proposed, which can be divided into two steps. First, the pixel correlation between adjacent pixels is retrieved based on patch-weighted distance, and then the pixel correlation is used to replace the influence of neighboring information in fuzzy algorithms, thereby enhancing the robustness. Experiments on simulated, natural and medical images illustrate that the proposed schema outperforms other fuzzy clustering algorithms.

     

一种直接高效的自然场景汉字逼近定位方法

为了提高经典目标检测算法对自然场景文本定位的准确性,以及克服传统字符检测模型由于笔画间存在非连通性引起的汉字错误分割问题,提出了一种直接高效的自然场景汉字逼近定位方法。采用经典的EAST算法对场景图像中的文字进行检测。对初检的文字框进行调整使其更紧凑和更完整地包含文字,主要由提取各连通笔画成分、汉字分割和文字形状逼近三部分组成。矫正文字区域和识别文字内容。实验结果表明,提出的算法在保持平均帧率为3.1 帧/s的同时,对ICDAR2015、ICDAR2017-MLT和MSRA-TD500三个多方向数据集上文本定位任务中的F-score分别达到83.5%、72.8%和81.1%;消融实验验证了算法中各模块的有效性。在ICDAR2015数据集上的检测和识别综合评估任务中的性能也验证了该方法相比一些最新方法取得了更好的性能。     

A fast technique for image segmentation based on two Meta-heuristic algorithms

Image segmentation is a primary task in image processing which is widely used in object detection and recognition. Multilevel thresholding is one of the prominent technique in the field of image segmentation. However, the computational cost of multilevel thresholding increases exponentially as the number of threshold value increases, which leads to use of meta-heuristic optimization to find the optimal number of threshold. To overcome this problem, this paper investigates the ability of two nature-inspired algorithms namely: antlion optimisation (ALO) and multiverse optimization (MVO). ALO is a population-based method and mimics the hunting behaviour of antlions in nature. Whereas, MVO is based on the multiverse theory which depicts that there is over one universe exist. These two metaheuristic algorithms are used to find the optimal threshold values using Kapur’s entropy and Otsu’s between class variance function. They examine the outcomes of the proposed algorithm with other evolutionary algorithms based on cost value, stability analysis, feature similarity index (FSIM), structural similarity index (SSIM), peak signal to noise ratio (PSNR), computational time. We also provide Wilcoxon test which justify the response of these parameters. The experimental results showed that the proposed algorithm gives better results than other existing methods. It is noticed that MVO is faster than other algorithms. The proposed method is also tested on medical images to detect the tumor from MRI T1-weighted contrast-enhanced brain images.

     

Integrated features and GMM Based Hand Detector Applied to Character Recognition System under Practical Conditions

Detection of bare-hand under non-ideal conditions is a challenging task. Most of the existing hand detection systems are developed under limited environmental constraints. In this study, a robust two-level bare-hand detector is integrated with a 58 keyboard characters recognition model. At first, the Gaussian mixture model (GMM) based foreground detector is used to segment the region of interest (ROI), which is further classified using Color-texture and texture based models to detect the actual fist. The detected hand is tracked using modified Kanade–Lucas–Tomasi (KLT) tracker to generate the required trajectory points of the character. The feature space for character recognition consists of existing features and three new features, namely, Local Geometrical Area Ratio (LGAR), Area of two halves (ATH), Curve-Area feature (CAF) that are extracted from the trajectory points. Feature space is optimized using statistical analysis algorithms. Multi-factor analysis of individual character subsets such as alphabets, numbers, ASCII characters, etc., are carried out using multiple conventional classifiers along with Support vector machine (SVM), extreme learning machine (ELM), artificial neural network (ANN), and proposed Neuro-fuzzy classifiers. The proposed GMM based motion detection method achieves an accuracy of 100% during the segmentation of ROI, followed by an increase of 46.77% in the accuracy of two-level hand detection under non-ideal conditions. Maximum accuracy of 58 character system using proposed features and ANN classifier is observed to be 92.56%.

     

Stacked Gated Recurrent Unit Classifier with CT Images for Liver Cancer Classification

Liver cancer is one of the major diseases with increased mortality in recent years, across the globe. Manual detection of liver cancer is a tedious and laborious task due to which Computer Aided Diagnosis (CAD) models have been developed to detect the presence of liver cancer accurately and classify its stages. Besides, liver cancer segmentation outcome, using medical images, is employed in the assessment of tumor volume, further treatment plans, and response monitoring. Hence, there is a need exists to develop automated tools for liver cancer detection in a precise manner. With this motivation, the current study introduces an Intelligent Artificial Intelligence with Equilibrium Optimizer based Liver cancer Classification (IAIEO-LCC) model. The proposed IAIEO-LCC technique initially performs Median Filtering (MF)-based pre-processing and data augmentation process. Besides, Kapur’s entropy-based segmentation technique is used to identify the affected regions in liver. Moreover, VGG-19 based feature extractor and Equilibrium Optimizer (EO)-based hyperparameter tuning processes are also involved to derive the feature vectors. At last, Stacked Gated Recurrent Unit (SGRU) classifier is exploited to detect and classify the liver cancer effectively. In order to demonstrate the superiority of the proposed IAIEO-LCC technique in terms of performance, a wide range of simulations was conducted and the results were inspected under different measures. The comparison study results infer that the proposed IAIEO-LCC technique achieved an improved accuracy of 98.52%.     

基于注意力机制和辅助任务的语义分割算法

为提高网络模型低层特征的离散度和语义分割算法的性能,以全卷积神经网络作为基础模型,提出一种基于辅助损失、边缘检测辅助任务和注意力机制的语义分割算法。通过重新设计网络模型的辅助损失分支,使网络低层特征编码更多语义信息。在多任务学习中,选择边缘检测作为辅助任务,基于注意力机制设计边缘检测的辅助任务分支,使网络模型更关注物体的形状和边缘信息。在此基础上,将基础模型、辅助损失分支、辅助任务分支集成构造为语义分割模型。在VOC2012数据集上的实验结果表明,该算法的平均交并比为71.5%,相比基础模型算法提高了6个百分点。     

Brain tumor detection based on hybrid deep neural network in MRI by adaptive squirrel search optimization

In the medical field, image segmentation is a paramount and challenging task. The head and vertebral column make up the central nervous system (CNS), which control all the paramount functions. These include thinking, speaking, and gestures. The uncontrolled growth in the CNS can affect a person’s thinking of communication or movement. The tumor is known as the uncontrolled growth of cells in brain. The tumor can be recognized by MRI image. Brain tumor detection is mostly affected with inaccurate classification. This proposed work designed a novel classification and segmentation algorithm for the brain tumor detection. The proposed system uses the Adaptive fuzzy deep neural network with frog leap optimization to detect normality and abnormality of the image. Accurate classification is achieved with error minimization strategy through our proposed method. Then, the abnormal image is segmented using adaptive flying squirrel algorithm and the size of the tumor is detected, which is used to find out the severity of the tumor. The proposed work is implemented in the MATLAB simulation platform. The proposed work Accuracy, sensitivity, specificity, false positive rate and false negative rate are 99.6%, 99.9%, 99.8%, 0.0043 and 0.543, respectively. The detection accuracy is better in our proposed system than the existing teaching and learning based algorithm, social group algorithm and deep neural network.

     

基于DenseNet-BC网络的皮肤镜下皮肤损伤分割

针对皮肤病变图像边界分割不准确的问题,提出了一种改进的稠密卷积网络(DenseNet-BC)皮肤损伤分割算法。首先,改变传统算法层与层之间的连接方式,通过密集连接使得所有层都能直接访问从原始输入信号到损失函数的梯度,让图像特征信息得到最大化的流动。其次,为降低参数数量与网络的计算量,在瓶颈层和过渡层中采用小卷积核对输入特征图的通道数进行减半操作。将DenseNet-BC算法与VGG-16、Inception-v3以及ResNet-50等算法在ISIC 2018 Task 1皮肤病变分割数据集上进行性能比较。实验结果表明,DenseNet-BC算法的病变分割准确率为0.975,Threshold Jaccard为0.835,分割准确率较其他算法提升显著,是一种有效的皮损分割算法。     

基于YOLOv5和FCN-DenseNet水下图像多目标语义分割算法

带视觉系统的水下机器人作业离不开对水下目标准确的分割, 但水下环境复杂, 场景感知精度和识别精度不高等问题会严重影响目标分割算法的性能. 针对此问题本文提出了一种综合YOLOv5和FCN-DenseNet的多目标分割算法. 本算法以FCN-DenseNet算法为主要分割框架, YOLOv5算法为目标检测框架. 采用YOLOv5算法检测出每个种类目标所在位置; 然后输入针对不同类别的FCN-DenseNet语义分割网络, 实现多分支单目标语义分割, 最后融合分割结果实现多目标语义分割. 此外, 本文在Kaggle竞赛平台上的海底图片数据集上将所提算法与PSPNet算法和FCN-DenseNet算法两种经典的语义分割算法进行了实验对比. 结果表明本文所提的多目标图像语义分割算法与PSPNet算法相比, 在MIoU和IoU指标上分别提高了14.9%和11.6%; 与FCN-DenseNet算法在MIoU和IoU指标上分别提高了8%和7.7%, 更适合于水下图像分割.     

Hybrid multi-GPU computing: accelerated kernels for segmentation and object detection with medical image processing applications

In the last two decades, we have seen an amazing development of image processing techniques targeted for medical applications. We propose multi-GPU-based parallel real-time algorithms for segmentation and shape-based object detection, aiming at accelerating two medical image processing methods: automated blood detection in wireless capsule endoscopy (WCE) images and automated bright lesion detection in retinal fundus images. In the former method we identified segmentation and object detection as being responsible for consuming most of the global processing time. While in the latter, as segmentation was not used, shape-based object detection was the compute-intensive task identified. Experimental results show that the accelerated method running on multi-GPU systems for blood detection in WCE images is on average 265 times faster than the original CPU version and is able to process 344 frames per second. By applying the multi-GPU framework for bright lesion detection in fundus images we are able to process 62 frames per second with a speedup average 667 times faster than the equivalent CPU version.     

深度学习多部位病灶检测与分割

目的 多部位病灶具有大小各异和类型多样的特点,对其准确检测和分割具有一定的难度。为此,本文设计了一种2.5D深度卷积神经网络模型,实现对多种病灶类型的计算机断层扫描（computed tomography,CT）图像的病灶检测与分割。方法 利用密集卷积网络和双向特征金字塔网络组成的骨干网络提取图像中的多尺度和多维度信息,输入为带有标注的中央切片和提供空间信息的相邻切片共同组合而成的CT切片组。将融合空间信息的特征图送入区域建议网络并生成候选区域样本,再由多阈值级联网络组成的Cascade R-CNN（region convolutional neural networks）筛选高质量样本送入检测与分割分支进行训练。结果 本文模型在DeepLesion数据集上进行验证。结果表明,在测试集上的平均检测精度为83.15%,分割预测结果与真实标签的端点平均距离误差为1.27 mm,直径平均误差为1.69 mm,分割性能优于MULAN（multitask universal lesion analysis network for joint lesion detection,tagging and segmentation）和Auto RECIST（response evaluation criteria in solid tumors）,且推断每幅图像平均时间花费仅91.7 ms。结论 对于多种部位的CT图像,本文模型取得良好的检测与分割性能,并且预测时间花费较少,适用病变类别与DeepLesion数据集类似的CT图像实现病灶检测与分割。本文模型在一定程度上能满足医疗人员利用计算机分析多部位CT图像的需求。     

Beyond document object detection: instance-level segmentation of complex layouts

Information extraction is a fundamental task of many business intelligence services that entail massive document processing. Understanding a document page structure in terms of its layout provides contextual support which is helpful in the semantic interpretation of the document terms. In this paper, inspired by the progress of deep learning methodologies applied to the task of object recognition, we transfer these models to the specific case of document object detection, reformulating the traditional problem of document layout analysis. Moreover, we importantly contribute to prior arts by defining the task of instance segmentation on the document image domain. An instance segmentation paradigm is especially important in complex layouts whose contents should interact for the proper rendering of the page, i.e., the proper text wrapping around an image. Finally, we provide an extensive evaluation, both qualitative and quantitative, that demonstrates the superior performance of the proposed methodology over the current state of the art.

     

Pathological lung segmentation based on random forest combined with deep model and multi-scale superpixels

Accurate segmentation of lungs in pathological thoracic computed tomography (CT) scans plays an important role in pulmonary disease diagnosis. However, it is still a challenging task due to the variability of pathological lung appearances and shapes. In this paper, we proposed a novel segmentation algorithm based on random forest (RF), deep convolutional network, and multi-scale superpixels for segmenting pathological lungs from thoracic CT images accurately. A pathological thoracic CT image is first segmented based on multi-scale superpixels, and deep features, texture, and intensity features extracted from superpixels are taken as inputs of a group of RF classifiers. With the fusion of classification results of RFs by a fractional-order gray correlation approach, we capture an initial segmentation of pathological lungs. We finally utilize a divide-and-conquer strategy to deal with segmentation refinement combining contour correction of left lungs and region repairing of right lungs. Our algorithm is tested on a group of thoracic CT images affected with interstitial lung diseases. Experiments show that our algorithm can achieve a high segmentation accuracy with an average DSC of 96.45% and PPV of 95.07%. Compared with several existing lung segmentation methods, our algorithm exhibits a robust performance on pathological lung segmentation. Our algorithm can be employed reliably for lung field segmentation of pathologic thoracic CT images with a high accuracy, which is helpful to assist radiologists to detect the presence of pulmonary diseases and quantify its shape and size in regular clinical practices.

     

基于改进Mask R-CNN的生活垃圾检测

近年来,人们对于垃圾的分类与回收越来越重视,但垃圾分类耗费了大量的人力和物力且分拣效率低下。针对基于矩形边界框的垃圾检测方法在多分类环境下效果不够理想等问题,提出了一种基于改进Mask R-CNN算法的生活垃圾检测模型。该模型摒弃了传统的ResNet,采用改进的ResNeXt101 作为主干网络进行特征提取,提高了目标检测的准确率和背景边界线分割的精确度。实验结果表明,与传统的Mask R-CNN算法相比,本文模型的mAP为91.1%,提升了2.35%;与当前流行的目标检测模型进行了对比,本文模型的分类准确率和分割精确度均表现优异,表明了所提模型在垃圾检测任务中的可行性与有效性。     

Fine-grained histopathological cell segmentation through residual attention with prior embedding

In the task of histopathological cell segmentation, traditional algorithms struggle with cell edge processing, which leads to the blurring of cell edges. To strengthen the ability to learn the features of cell edges, this paper develops a novel deep neural network for robust and fine-grained cell segmentation. The proposed deep model mines global and local features by multiscale convolution and dilated convolution. Subsequently, the residual attention module is introduced in the third to fifth layers of the encoder; this module assigns a group of weight coefficients to all the deep features to boost the segmentation performance. In addition, to further improve the quality of the features in the decoder, we first introduce the strategy of U-Net for the extraction of prior information, where we filter the fused features and compress the features by using the prior information and the filtered features again to integrate more semantic information into the feature refinement in the decoding process. We tested the model on three public data sets: Multiorgan Nucleus Segmentation (MoNuSeg) (Dice 94.9%), Triple Negative Breast Cancer (TNBC) (Dice 95.4%) and Data Science Bowl (Dice 98.2%). Extensive experiments demonstrate the superior performance of our proposed method in comparison with that of state-of-the-art models; our method can effectively identify cell edges to produce fine-grained segmentation results.

     

A localization method for stagnant water in city road traffic image

Stagnant water on roads has always been a major cause of traffic jams and accidents. Traditional urban waterlogging monitoring and warning system is mainly based on a large amount of historical data and predictive network, which has low accuracy and weak generalization ability. Considering the deep neural network algorithms have demonstrated strong capabilities in computer vision tasks such as object detection, we aim to apply them to road stagnant water detection. In this paper, a novel automatic stagnant water localization method under weak supervision based on visual image is proposed. First, the template matching method is applied to extract road information from the traffic image. Then, due to the complexity of data annotation, we locate stagnant water in image based on Class Activation Maps (CAM) mechanism, which is a weakly supervised method. The detection model consists of the ResNet-18 and the Grad-CAM++ mechanism. Finally, based on the heat map and template, we set a suitable threshold to segment stagnant water area in image. In the experiments, the precision and recall for road stagnant water classification by the proposed model are 99.39% and 99.60%, while the Intersection over Union (IoU) for stagnant water area segmentation is up to 63%. These show that our method is effective for road stagnant water localization.