2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)最新文献

{"title":"Automated Segmentation of Nucleus, Cytoplasm and Background of Cervical Cells from Pap-smear Images using a Trainable Pixel Level Classifier","authors":"W. Wasswa, J. Obungoloch, A. H. Basaza-Ejiri, Andrew Ware","doi":"10.1109/AIPR47015.2019.9174599","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174599","url":null,"abstract":"Cervical cancer ranks as the fourth most prevalent cancer affecting women worldwide and its early detection provides the opportunity to help save life. Automated diagnosis of cervical cancer from pap-smear images enables accurate, reliable and timely analysis of the condition’s progress. Cell segmentation is a fundamental aspect of successful automated pap-smear analysis. In this paper, a potent approach for segmentation of cervical cells from a pap-smear image into the nucleus, cytoplasm and background using pixel level information is proposed. A number of pixels from the nuclei, cytoplasm and background are extracted from 100 images to form a feature vector which is trained using noise reduction, edge detection and texture filters to produce a pixel level classifier. Comparison of the segmented images’ nucleus and cytoplasm parameters (nucleus area, longest diameter, roundness, perimeter and cytoplasm area, longest diameter, roundness, perimeter) with the ground truth image features yielded average percentage errors of 0.14, 0.28, 0.03, 0.30, 0.15, 0.25, 0.05 and 0.39 respectively. Validation of the pixel classifier with 10fold cross-validation yielded pixel classification accuracy of 98.50%, 97.70% and 98.30% with Fast Random Forest, Naïve Bayes and J48 classification methods respectively. Comparison of the segmented nucleus and cytoplasm with the ground truth nucleus and cytoplasm segmentations resulted into a Zijdenbos similarity index greater than 0.9321 and 0.9639 for nucleus and cytoplasm segmentation respectively. The results indicated that the proposed pixel level segmentation classifier was able to extract the nucleus and cytoplasm regions accurately and worked well even though there was no significant contrast between the components in the image. The results from cross-validation and test set evaluation imply that the classifier can segment cells outside the training dataset with high precision. Choosing an appropriate feature vector for training the classifier was a great challenge and a novel task in the proposed approach. As a result, good segmentation of the nucleus and cytoplasm was attained. Given the accuracy of the classifier in segmenting the nucleus, which plays an important role in cervical cancer diagnosis, the classifier can be adopted in systems for automated diagnosis of cervical cancer from pap-smear images.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114286603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sensitivity of Multiview 3D Point Cloud Reconstruction to Compression Quality and Image Feature Detectability","authors":"Ke Gao, Shizeng Yao, H. Aliakbarpour, S. Agarwal, G. Seetharaman, K. Palaniappan","doi":"10.1109/AIPR47015.2019.9174580","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174580","url":null,"abstract":"In this paper we evaluate the quality of feature detection and 3D reconstruction on a Wide Area Motion Imagery (WAMI) sequence with increasing JPEG compression ratio. Feature detection is critical for computer vision tasks such as 3D reconstruction. For some 3D reconstruction approaches, the quality of a 3D model relies upon consistent detection of the same feature points over consecutive frames in an image sequence. Since the performance of feature detectors is highly sensitive to compression artifacts, we evaluate the influence of image quality on feature detection accuracy. Many datasets (e.g. WAMI) use JPEG compression to decrease the data storage and network bandwidth utilization while attempting to preserve image quality by adaptively adjusting the compression ratio. Consequently, it is important to understand the impact of JPEG compression on the quality of feature detection in 2D space and the subsequent 3D reconstruction results. We design and perform two evaluation procedures on the WAMI sequence. We use structure tensor to detect feature points on an image sequence with increasing JPEG compression ratio (10:1, 15:1, 20:1, 30:1, 40:1, 100:1, and 150:1). Compression ratio of 10:1 is used as the baseline (groundtruth). First we compare the feature points from images of different qualities with the groundtruth features and evaluate them on pixel level in 2D space. After that, a 3D model in the form of point cloud is generated from each set of feature points and compared with the groundtruth point cloud. We provide quantitative and visualized results for the evaluation.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126537547","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Object Counting using KAZE Features Under Different Lighting Conditions for Inventory Management","authors":"Teena Sharma, Astha Jain, N. Verma, S. Vasikarla","doi":"10.1109/AIPR47015.2019.9174578","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174578","url":null,"abstract":"Inventory management in an automated industrial environment is the foremost requirement in order to shorten the gap between demand and supply. It comprises of object identification, localization, and counting. This paper introduces an approach for object counting in automated inventory management using KAZE features under different lighting conditions. Firstly, the prototype image and real-time inventory feed as the scene image are captured for detection of KAZE features. The detected features in the prototype image are subjected to density based scanning clustering algorithm. The KAZE features of each cluster obtained in the prototype image are mapped with the KAZE features of inventory feed. The mapped features in inventory feed are again subjected to density based scanning clustering algorithm. The clusters obtained in the inventory feed are then processed by Homography transform. Homography transform generates the predictions for object locations by projecting prototype corners in the inventory feed. The Homography transform projection results in rectangular box polygons in the inventory feed for the tentative location of prototype instances. Since there may be multiple predictions for a single object instance, the predicted object locations are integrated by density based scanning clustering algorithm to the centroids of these rectangular box polygons. It provides the exact location of prototype instances. Finally, the count is obtained. The graphical user interface for inventory management is also designed which exhibits user-friendly attributes. The proposed approach has also been compared with the previously developed approaches of object counting in automated inventory management. The experimental results state that the proposed approach outperforms the existing ones in the presence of different lighting conditions such as low-light or dim-light and bright light.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130275939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SRC3: A Video Dataset for Evaluating Domain Mismatch","authors":"Jonathan Sato, Chelsea Mediavilla, C. Ward, S. Parameswaran","doi":"10.1109/AIPR47015.2019.9174589","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174589","url":null,"abstract":"In this paper we introduce new video datasets to investigate the gaps between synthetic and real imagery in object detection and depth estimation. Currently, synthetic image datasets with real-world counterparts largely focus on computer vision applications for autonomous driving in unconstrained environments. The high scene complexity of such datasets pose challenges for systematic studies of domain disparities. We aim to create a set of paired datasets to study the discrepancies between the two domains in a more controlled setting. To this end, we have created Synthetic-Real Counterpart 3 (SRC3), which contains multiple datasets with varying levels of scene and object complexity. These versatile datasets span multiple environments and consist of ground-truthed, real-world, and synthetic videos generated using a gaming engine. In addition to the dataset, we present an in-depth analysis and provide comparison benchmarks of these datasets using state-of-the-art detection algorithms. Our results show contrasting performance during cross-domain testing due to differences in image quality and statistics, indicating a need for domain adapted datasets and models.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127068017","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Image Representations for Occluded Face Recognition via Reference Conditioned Low-Rank projection","authors":"Shibashish Sen, Manikandan Ravikiran","doi":"10.1109/AIPR47015.2019.9174567","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174567","url":null,"abstract":"Deep learning in face recognition is widely explored in recent times due to its ability to produce state-of-the-art results and availability of large public datasets. While recent deep learning approaches involving margin loss based image representations produce 99% accuracy across benchmarks, none of these studies focus explicitly on occluded face verification. Further, in real world scenarios, there is a need for efficient methods that cater to the cases of occlusion of faces with hats, scarves, goggle or sometimes exaggerated facial expression. Moreover, with face verification gathering traction in mainstream real-time embedded applications of surveillance, the proposed approaches need to be highly accurate. In this paper, we revisit the same through a large-scale study involving multiple synthetically created goggle-occluded face datasets using multiple state-of-the-art face representations. Through this study, we identify that occlusion in faces results in non-isotropic face representations in feature space which results in a drop in performance. Therefore, we propose an approach to enhance existing face representations by learning reference conditioned Low-Rank projections (RCLP), which can create isotropic representations thereby improving face recognition. We benchmark the developed approach over synthetically goggled versions of LFW, CFP-FP, ATT, FEI, Georgia Tech and Essex University face databases with representations from ResNet-ArcFace, VGGFace, MobilefaceNet-ArcFace LightCNN resulting in a total of 100 + experiments where we achieve improvements in the accuracy-rate across all with a maximum of 4% on FEI dataset. Finally, to validate the approach in a realistic scenario, we additionally present results over our internal face verification dataset of 1k images and confirm that the proposed approach only shows positive results without degrading existing baseline performance.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126076658","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient Passive Sensing Monocular Relative Depth Estimation","authors":"Alex Yang, G. Scott","doi":"10.1109/AIPR47015.2019.9174573","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174573","url":null,"abstract":"We propose a method to perform monocular relative depth perception using a passive visual sensor. Specifically, the proposed method makes depth estimation with a superpixel based regression model based on features extracted by a deep convolutional neural network. We have established and conducted an analysis of the key components required to create a high-efficiency pipeline to solve the depth estimation problem with superpixel-level regression and deep learning. The key contributions of our method compared to prior works are as follows. First, we have drastically simplified the depth estimation model while attaining near state-of-the-art prediction performance, through two important optimizations: the idea of the depth estimation model is completely based on superpixels that very effectively reduces the dimensionality; additionally, we exploited the scale invariant mean squared error loss function which incorporates a pairwise term with linear time complexity. Additionally, we have developed optimizations of the superpixel feature extraction, that leverage GPU computing to achieve real-time performance (over 50fps during training) Furthermore, this model does not perform up-sampling, which avoids many issues and difficulties that one would otherwise have to deal with. To perpetuate future research in this area we have created a synchronized multiple-view depth estimation training dataset that is available to the public.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126738318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluation of Generative Adversarial Network Performance Based on Direct Analysis of Generated Images","authors":"Shuyue Guan, M. Loew","doi":"10.1109/AIPR47015.2019.9174595","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174595","url":null,"abstract":"Recently, a number of papers have addressed the theory and applications of the Generative Adversarial Network (GAN) in various fields of image processing. Fewer studies, however, have directly evaluated GAN outputs. Those that have been conducted focused on using classification performance and statistical metrics. In this paper, we consider a fundamental way to evaluate GANs by directly analyzing the images they generate, instead of using them as inputs to other classifiers. We consider an ideal GAN according to three aspects: 1) Creativity: non-duplication of the real images. 2) Inheritance: generated images should have the same style, which retains key features of the real images. 3) Diversity: generated images are different from each other. Based on the three aspects, we have designed the Creativity-Inheritance-Diversity (CID) index to evaluate GAN performance. We compared our proposed measures with three commonly used GAN evaluation methods: Inception Score (IS), Fréchet Inception Distance (FID) and 1-Nearest Neighbor classifier (1NNC). In addition, we discuss how the evaluation could help us deepen our understanding of GANs and improve their performance.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115801699","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep Nets Spotlight Illegal, Unreported, Unregulated (IUU) Fishing","authors":"Darrell L. Young","doi":"10.1109/AIPR47015.2019.9174577","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174577","url":null,"abstract":"The need for increased global surveillance and enforcement efforts to combat Illegal, Unreported, Unregulated (IUU) fishing is well known. This paper describes the current research status in developing a novel technique of associating Automated Identification System (AIS) anti-collision messages to satellite vessel detects. Each detected ship image has a wealth of information which allows development of dark ship tracking and identification. A dark ship is a ship that is not broadcasting AIS. Ships involved in illegal activities often disable their AIS transmitter to avoid detection by authorities. Dark ship tracking and identification uses a deep similarity metrics to compare current and previous observations. If any of the previous observations have an identity, e.g. a known vessel on the international IUU watch-list, then the probability of its involvement in illegal activity is increased. Additional indicators of IUU activity such as frequent flag changes are combined in a probabilistic evaluation of accumulated evidence using local laws, rules, and regulations to render IUU assessments using commercially available imagery and data sources.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117164645","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implicit Land Use Mapping Using Social Media Imagery","authors":"Connor Greenwell, Scott Workman, Nathan Jacobs","doi":"10.1109/AIPR47015.2019.9174570","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174570","url":null,"abstract":"Land use classification is a central remote sensing task with a broad range of applications. Typically this is represented as a supervised learning problem, the first step of which is to develop a taxonomy of discrete labels. However, such categories are restricted in the range of uses they can convey and arbitrary decisions are often required when defining the categories. Instead, we argue that the abstract notion of land use can be indirectly characterized by the types and quantities of common objects found in an area. To capture the presence of such objects, we propose an implicit approach to defining and estimating land use that relies on sparsely distributed social media imagery but retains the benefits of dense coverage provided by satellite imagery. Our method is formulated as a convolutional neural network that operates on satellite imagery and outputs a probability distribution over quantities of objects common in social media imagery at that location. We show that the learned feature representation is discriminative for existing land use categories.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129340854","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Parasite Detection in Thick Blood Smears Based on Customized Faster-RCNN on Smartphones","authors":"Feng Yang, Hang Yu, K. Silamut, R. Maude, Stefan Jaeger, Sameer Kiran Antani","doi":"10.1109/AIPR47015.2019.9174565","DOIUrl":"https://doi.org/10.1109/AIPR47015.2019.9174565","url":null,"abstract":"Malaria is a worldwide life-threatening disease. The gold standard for malaria diagnosis is microscopy examination, which includes thick blood smears to detect the presence of parasites and thin blood smears to differentiate the development stages of parasites. Microscopy examination is of low cost but is time consuming and error-prone. Therefore, the development of an automated parasite detection system for malaria diagnosis in thick blood smears is an important research goal, especially in resource-limited areas. In this paper, based on a customized Faster-RCNN model, we develop a machine-learning system that can automatically detect parasites in thick blood smear images on smartphones. To make Faster-RCNN more efficient for small object detection, we split an input image of $4032 times 3024 times3$ pixels into small blocks of $252 times 189 times3$ pixels, and then train the FasterRCNN model with the small blocks and corresponding parasite annotations. Moreover, we customize the convolutional layers of Faster-RCNN with four convolutional layers and two maxpooling layers to extract features according to the input image size and characteristics. We perform experiments on 2967 thick blood smear images from 200 patients, including 1819 images from 150 patients who are infected with parasites. The customized FasterRCNN model is first trained on small image blocks from 120 patients, including 90 infected patients and 30 normal patients, and then tested on the remaining 80 patients. For testing, we also split each input image into small blocks of $252 times 189 times3$ pixels that are screened by our trained Faster-RCNN model to detect parasite coordinates, which are then re-projected into the original image space. Detection rates of our system on image level and patient level are 96.84% and 96.81%, respectively.","PeriodicalId":167075,"journal":{"name":"2019 IEEE Applied Imagery Pattern Recognition Workshop (AIPR)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127181167","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}