Xiaoliang Hao, Qianqian Yang, Yinfeng Xia, B. Yin, Sifan Peng
{"title":"Feature Reaggregation Network for Crowd Counting","authors":"Xiaoliang Hao, Qianqian Yang, Yinfeng Xia, B. Yin, Sifan Peng","doi":"10.1145/3441250.3441260","DOIUrl":"https://doi.org/10.1145/3441250.3441260","url":null,"abstract":"In this paper, we propose a novel end-to-end network named Feature Reaggregation Network (FRNet) for crowd counting, which focuses on fusing the multi-scale features in the hierarchy for generating high-quality density maps. Two level and three level feature reaggregation modules are developed between the backbone network and the next feature extraction modules in the hierarchy so that the low-layer spatial feature and the high-layer semantic information can be multi-combined by element-wise addition. In addition, these two modules are placed behind the backbone network to extract global features. Furthermore, we also introduce the extra deformable mixture regression module which donates deformable convolution to extract features so that we can generate the high-quality estimated density map. We have evaluated our experiments on three popular crowd counting datasets (ShanghaiTech, UCF_CC_50 and UCF_QNRF datasets), and the experiments demonstrate that the superiority of the proposed method over the other excellent approaches.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126797836","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":"CMRS: Towards Intelligent Recommendation for Choosing College Majors","authors":"Yan Meng, Min Fu","doi":"10.1145/3441250.3441272","DOIUrl":"https://doi.org/10.1145/3441250.3441272","url":null,"abstract":"The choice of college majors plays a critical role in determining the future opportunities and earnings of college graduates. Students make their college major decisions due to heterogeneous tastes and abilities, and these individual characteristics determine students’ performance in campus. However, existing methods on major recommendation have several limitations such as incomplete features and characteristics. In this paper we propose a machine-learning based college major recommendation approach, called CMRS, for prospective freshmen to choose desired majors. Our approach compared different state-of-art machine learning models to figure out the best one as our recommendation approach, namely Support Vector Machine (SVM), Decision Tree (DT), Naive Bayes (GNB), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) as well as collaborative filtering (CF). For this purpose we classified a dataset of more than 2000 college students from different majors in China. We compared the mean overall classification accuracies as well as standard deviation. The experimental results demonstrate that Random Forest exhibited superior results to other considered models in terms of overall accuracy and robustness, which produces the optimal college major recommendation accuracy of 97.87% and f-score of 96.60%.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124451544","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}
T. Panggabean, M. E. Simaremare, Rusmina Siahaan, Chandro Pardede, Wiwin Putri Gurning
{"title":"Another Parallelism Technique of GLCM Implementation Using CUDA Programming","authors":"T. Panggabean, M. E. Simaremare, Rusmina Siahaan, Chandro Pardede, Wiwin Putri Gurning","doi":"10.1145/3441250.3441251","DOIUrl":"https://doi.org/10.1145/3441250.3441251","url":null,"abstract":"In pattern recognition and image processing feature extraction is one of essential parts as it facilitates the subsequent learning and generalization steps. Feature extraction aims to get the special characteristics of an object that can be applied to produce the desired information. One of commonly used algorithm for feature extraction is Gray Level Co-occurrence Matrix (GLCM). Though this algorithm produces a good result, the twelve Haralick texture features requires high computation if CPUs are used. The emergence of the Graphics Processing Unit (GPU) helps cope with all that requires high computation and running GLCM calculations in GPU is just a matter of seconds. To benefit greatly from thousands of CUDA cores, one should be able to appropriately implement GLCM algorithm. GPU for GLCM had been done in several studies and each offered various improvement in computation times. In this research, we propose another technique of implementing GLCM algorithm and it shows that the computation time of the twelve Haralick texture features is 450 times faster than using CPU.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115111809","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":"Shrinkage Method for Categorical Explanatory Variables","authors":"Siyao Zhang","doi":"10.1145/3441250.3441275","DOIUrl":"https://doi.org/10.1145/3441250.3441275","url":null,"abstract":"Significantly more immigrant students are served by US public schools. To provide more efficient support to those immigrant students, we need to investigate the current situation of those immigrant students and graduation rate is used as a reference of their situation. In this article, fused lasso model is used as a shrinkage method for categorical explanatory variables. Data from graduation rate of New York state are considered. The result is given that students with better English skills and from Asian/Pacific Islander are more likely to have a higher graduation rate, which may suggest language support can be an efficient way to help those immigrant students.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131017813","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":"Key Frame Extraction of Evidence Video Based on YOLO Algorithm","authors":"Jinming Min, Jihong Zheng, Gareth J. F. Jones","doi":"10.1145/3441250.3441278","DOIUrl":"https://doi.org/10.1145/3441250.3441278","url":null,"abstract":"For the massive forensic video data, if we watch and intercept the key information manually, it will take a lot of time and delay the opportunity of obtaining useful evidence. For real-time processing the data with high accuracy, the computer automatically extracts the video frames with high-value content (including the image frames of specified types of aircraft, ships, and military uniforms in our experiments) through automatic video analysis. This method achieves the preliminary screening requirements for the key-frames, thus greatly reducing the time required for manually processing a large number of images. In this work, we propose a new method to utilize YOLO algorithm to conduct key-frame extraction in processing the video data.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125227778","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":"Context-based Feature Fusion Network for Object Detection","authors":"Yinfeng Xia, Yuqiang He, Xiaoliang Hao, B. Yin","doi":"10.1145/3441250.3441259","DOIUrl":"https://doi.org/10.1145/3441250.3441259","url":null,"abstract":"For object detection, the method based on multi-scale feature fusion are widely used to improve performance. However, performance improvements are limited due to the absence of sufficient global context information and the presence of various background noises in low-level features. Regarding the above issues, we propose the Context Extraction Module (CEM) and Context Refinement Module (CRM). The CEM enables the network can learn abstract global context information from various receptive fields; the CRM enhances the discriminative ability of low-level features representations by using the attention mechanism. In order to evaluate the effectiveness of CEM and CRM, we built a new network called Context-based Feature Fusion Network (CFFN) based on FSSD (Feature Fusion Single Shot Multibox Detector) and conducted experiments. In the Pascal VOC 2007 test, the detection performance can achieve 81.4% mAP with the input size 320×320. In addition, our results in COCO also have greater advantages than conventional FSSD.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133208761","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":"Threshold Moving Approaches for Addressing the Class Imbalance Problem and their Application to Multi-label Classification","authors":"Xing Zhang, Hyukjun Gweon, S. Provost","doi":"10.1145/3441250.3441274","DOIUrl":"https://doi.org/10.1145/3441250.3441274","url":null,"abstract":"In this paper, simple threshold moving techniques are proposed for the class imbalance problem. Decision thresholds are adjusted to match the class distribution of the training data to that of the predicted outcomes for unseen data. The proposed approaches are applied to multi-label classification wherein each instance may simultaneously belong to more than one label. Experimental results on two data sets confirm that the performance of the well-known binary relevance method can be improved when combined with the proposed threshold moving techniques.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"75 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127731286","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":"A Transfer Learning Approach to Compound Facial Expression Recognition","authors":"Yuanlun Xie, Wenhong Tian, Tingsong Ma","doi":"10.1145/3441250.3441263","DOIUrl":"https://doi.org/10.1145/3441250.3441263","url":null,"abstract":"Recent advanced research shows that deep learning has great potential in facial expression recognition, specially, in the basic expression recognition field, many researchers have proposed lots of deep learning networks with excellent performance. As for compound facial expression, although it has great research value, there are only few researches existed, and their performance is not satisfactory. So, there is still a lot of work to be done to improve compound facial expression recognition. To address this problem, we introduce a transfer learning solution for compound expression recognition, where we design a fine-tuning structure combining with a VGG-19 pre-trained network. The VGG-19 pre-trained network is used to extract deep features of facial expression, and the fine-tuning network is used to obtain the category of the facial expression. The experimental results show that our approach improves the state-of-the-art deep learning method by about 5.31% in accuracy. As far as we know, the transfer learning approach is the first time to be applied to compound facial expression recognition based on CFEE dataset.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"27 Suppl 1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124574159","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}
Jie Liu, Rencan Nie, Yanglin Deng, Huaguang Li, Tianyu Mu
{"title":"An Image Fusion Method Based on Weighted Nonnegative Matrix Factorization and Infrared Fidelity","authors":"Jie Liu, Rencan Nie, Yanglin Deng, Huaguang Li, Tianyu Mu","doi":"10.1145/3441250.3441267","DOIUrl":"https://doi.org/10.1145/3441250.3441267","url":null,"abstract":"In image fusion, the most desirable information is obtained from multiple images of the same scene and fused to produce a composite image, which is more suitable for human visual perception and further image processing tasks. In this paper we propose a novel image fusion method based on weighted nonnegative matrix factorization and infrared fidelity. Firstly, the luminance contrast of infrared and visible images is extracted to obtain the luminance saliency maps, whose weight is estimated as the weighted term of the non-negative matrix factorization. Then norm is used to constrain the pixel intensity distance between the fusion result of WNMF and the infrared image, and update rules of the algorithm are obtained according to the gradient descent method. Finally the fused image is obtained. Compared with the three most advanced methods, experimental results indicate that this method is superior to previous methods in visual effects and quantifiably evaluate.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127650198","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":"Smooth Test for Meta-Elliptical Distribution of Innovations in the Vector Autoregressive Model","authors":"Yan Su","doi":"10.1145/3441250.3441276","DOIUrl":"https://doi.org/10.1145/3441250.3441276","url":null,"abstract":"This paper presents a new approach to testing the meta-elliptical distribution of the innovations in the vector autoregressive (VAR) model. The test is based on the smooth test for uniformity on the surface of a unit sphere. The asymptotic null distribution of the transformed residuals from the VAR model is obtained. A bootstrapping algorithm is given to estimate the critical values of the test for a wide range of sample sizes. Furthermore, the components of the smooth test can be used to detect specific types of departures from the null hypothesis.","PeriodicalId":205484,"journal":{"name":"Proceedings of the 4th International Conference on Advances in Image Processing","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125962221","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}