Pub Date : 2023-02-14DOI: 10.1109/IPRIA59240.2023.10147188
Alireza Sadeghi, Hassan Khutanlou
The hippocampus is an important part of the human brain that is damaged in some diseases such as Alzheimer's, schizophrenia, and epilepsy. This paper presents a new method in hippocampus segmentation which is applicable in the early diagnosis of mentioned diseases. This method has introduced a two-section model to detect the hippocampus region in brain MR images. In the first section, the location of the hippocampus is roughly detected using a U-Net neural network model, and then a fuzzy mask is created around the detected area using a fuzzy function. In the second section, this mask is applied to the brain images and a U-Net neural network is used to segment these masked images, which finally predicts the location of the hippocampus. The main advantage and idea of this method is the use of a pre-trained fuzzy mask, which increases the quality of segmentation. The proposed method in this research was trained and tested using the HARP dataset, which contains 135 T1-weighted MRI volumes and the proposed model reached 0.95 dice in the best case.
{"title":"Hippocampus segmentation in MR brain images using learned fuzzy mask and U-Net","authors":"Alireza Sadeghi, Hassan Khutanlou","doi":"10.1109/IPRIA59240.2023.10147188","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147188","url":null,"abstract":"The hippocampus is an important part of the human brain that is damaged in some diseases such as Alzheimer's, schizophrenia, and epilepsy. This paper presents a new method in hippocampus segmentation which is applicable in the early diagnosis of mentioned diseases. This method has introduced a two-section model to detect the hippocampus region in brain MR images. In the first section, the location of the hippocampus is roughly detected using a U-Net neural network model, and then a fuzzy mask is created around the detected area using a fuzzy function. In the second section, this mask is applied to the brain images and a U-Net neural network is used to segment these masked images, which finally predicts the location of the hippocampus. The main advantage and idea of this method is the use of a pre-trained fuzzy mask, which increases the quality of segmentation. The proposed method in this research was trained and tested using the HARP dataset, which contains 135 T1-weighted MRI volumes and the proposed model reached 0.95 dice in the best case.","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"29 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134035191","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}
Pub Date : 2023-02-14DOI: 10.1109/IPRIA59240.2023.10147170
Alireza Khatami, Ahmad Mahmoudi-Aznaveh
Measuring the perceptual similarity between two images is a long-standing problem. This assessment should mimic human judgments. Considering the complexity of the human visual system, it is challenging to model human perception. On the other hand, the recent low-level vision task approaches, mostly based on supervised deep learning, require an appropriate loss for the backward pass. The per-pixel loss, such as MSE and MAE, between the output of the network and the ground-truth images were among the first choices. More complicated and common similarity measures in which the error is computed in a hand-designed feature space are also employed. Furthermore, Deep Perceptual Similarity (DPS) metrics, where the similarity is measured in the deep feature space, also have promising results. This feature can be selected from a pre-trained or optimized model for the task at hand. Recently many studies have been conducted to thoroughly investigate DPS. In this research, we provide an in-depth analysis of the pros and cons of DPS in assessing the full reference quality assessment. In addition, to compare different similarity measures, we propose a metric which aggregates various desired factors. Based on our experiment, it can be concluded that perceptual similarity is not directly related to classification accuracy. It is discovered that the outliers mostly contain high-frequency elements. The code and complete outcomes described in results, can be found on: https://github.com/Alireza-Khatami/PerceptualQuality
{"title":"Deep perceptual similarity and Quality Assessment","authors":"Alireza Khatami, Ahmad Mahmoudi-Aznaveh","doi":"10.1109/IPRIA59240.2023.10147170","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147170","url":null,"abstract":"Measuring the perceptual similarity between two images is a long-standing problem. This assessment should mimic human judgments. Considering the complexity of the human visual system, it is challenging to model human perception. On the other hand, the recent low-level vision task approaches, mostly based on supervised deep learning, require an appropriate loss for the backward pass. The per-pixel loss, such as MSE and MAE, between the output of the network and the ground-truth images were among the first choices. More complicated and common similarity measures in which the error is computed in a hand-designed feature space are also employed. Furthermore, Deep Perceptual Similarity (DPS) metrics, where the similarity is measured in the deep feature space, also have promising results. This feature can be selected from a pre-trained or optimized model for the task at hand. Recently many studies have been conducted to thoroughly investigate DPS. In this research, we provide an in-depth analysis of the pros and cons of DPS in assessing the full reference quality assessment. In addition, to compare different similarity measures, we propose a metric which aggregates various desired factors. Based on our experiment, it can be concluded that perceptual similarity is not directly related to classification accuracy. It is discovered that the outliers mostly contain high-frequency elements. The code and complete outcomes described in results, can be found on: https://github.com/Alireza-Khatami/PerceptualQuality","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126049493","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}
Pub Date : 2023-02-14DOI: 10.1109/IPRIA59240.2023.10147177
Mahsa Mohamadi, Ako Bartani, F. Tab
The outdoor images are usually contaminated by atmospheric phenomena, which have effects such as low contrast, and poor quality and visibility. As the resulting dust phenomena is increasing day by day, improving the quality of dusty images as per-processing is an important challenge. To address this challenge, we propose a self-supervised method based on generative adversarial network. The proposed framework consists of two generators master and supporter which are trained in joint form. The master and supporter generators are trained using synthetic and real dust images respectively which their labels are generated in the proposed framework. Due to lack of real-world dusty images and the weakness of synthetic dusty image in the depth, we use an effective learning mechanism in which the supporter helps the master to generate satisfactory dust-free images by learning restore depth of Image and transfer its knowledge to the master. The experimental results demonstrate that the proposed method performs favorably against the previous dusty image enhancement methods on benchmark real-world duty images.
{"title":"Self-Supervised Dusty Image Enhancement Using Generative Adversarial Networks","authors":"Mahsa Mohamadi, Ako Bartani, F. Tab","doi":"10.1109/IPRIA59240.2023.10147177","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147177","url":null,"abstract":"The outdoor images are usually contaminated by atmospheric phenomena, which have effects such as low contrast, and poor quality and visibility. As the resulting dust phenomena is increasing day by day, improving the quality of dusty images as per-processing is an important challenge. To address this challenge, we propose a self-supervised method based on generative adversarial network. The proposed framework consists of two generators master and supporter which are trained in joint form. The master and supporter generators are trained using synthetic and real dust images respectively which their labels are generated in the proposed framework. Due to lack of real-world dusty images and the weakness of synthetic dusty image in the depth, we use an effective learning mechanism in which the supporter helps the master to generate satisfactory dust-free images by learning restore depth of Image and transfer its knowledge to the master. The experimental results demonstrate that the proposed method performs favorably against the previous dusty image enhancement methods on benchmark real-world duty images.","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121123625","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}
Pub Date : 2023-02-14DOI: 10.1109/IPRIA59240.2023.10147178
Marjan Mavaddat, M. Naderan, Seyyed Enayatallah Alavi
In the past, diagnosing pests has been a very important and challenging task for farmers, and ocular detection methods with the help of phytosanitary specialists, were time consuming, costly, and associated with human error. Today, in modern agriculture, diagnostic softwares by artificial intelligence can be used by farmers themselves with little time and cost. On the other hand, because diseases and pests of plants, especially rice leaves, are of different intensities and are similar to each other, automatic detection methods are more accurate and have less error. In this paper, two transfer learning methods for diagnosing rice leaf disease are investigated. The first method uses the CNN-based output of a pre-trained model and an appropriate classifier is added. In the second method, freezing the bottom layers, fine-tuning the weights in the last layers of the pre-trained network, and adding the appropriate classifier to the model are proposed. For this purpose, seven CNN models have been designed and evaluated. Simulation results show that four of these networks as: VGG16 network with fine tuning the last two layers, Inceptionv3 with fine tuning the last 12 layers, Resnet152v2 with fine tuning the last 5 and 6 layers reach 100% accuracy and an f1-score of 1. In addition, fewer number of layers in VGG16 network with 2-layers fine tuning consumes less memory and has faster response time. Also, our paper has a higher accuracy and less training time than similar papers.
{"title":"Classification of Rice Leaf Diseases Using CNN-Based Pre-Trained Models and Transfer Learning","authors":"Marjan Mavaddat, M. Naderan, Seyyed Enayatallah Alavi","doi":"10.1109/IPRIA59240.2023.10147178","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147178","url":null,"abstract":"In the past, diagnosing pests has been a very important and challenging task for farmers, and ocular detection methods with the help of phytosanitary specialists, were time consuming, costly, and associated with human error. Today, in modern agriculture, diagnostic softwares by artificial intelligence can be used by farmers themselves with little time and cost. On the other hand, because diseases and pests of plants, especially rice leaves, are of different intensities and are similar to each other, automatic detection methods are more accurate and have less error. In this paper, two transfer learning methods for diagnosing rice leaf disease are investigated. The first method uses the CNN-based output of a pre-trained model and an appropriate classifier is added. In the second method, freezing the bottom layers, fine-tuning the weights in the last layers of the pre-trained network, and adding the appropriate classifier to the model are proposed. For this purpose, seven CNN models have been designed and evaluated. Simulation results show that four of these networks as: VGG16 network with fine tuning the last two layers, Inceptionv3 with fine tuning the last 12 layers, Resnet152v2 with fine tuning the last 5 and 6 layers reach 100% accuracy and an f1-score of 1. In addition, fewer number of layers in VGG16 network with 2-layers fine tuning consumes less memory and has faster response time. Also, our paper has a higher accuracy and less training time than similar papers.","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"134 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114313390","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}
Pub Date : 2023-02-14DOI: 10.1109/IPRIA59240.2023.10147196
E. Afshar, Hassan Khotanlou, Elham Alighardash
Researchers have shown that 55% of concepts are conveyed through facial emotion and only 7% are conveyed by words and sentences, so facial expression plays an important role in conveying concepts in human communications. In recent years, due to the improvement of artificial neural networks, many studies have been conducted related to facial expression recognition. This paper presents a method based on ensemble classification using convolutional neural networks to recognize facial emotions. The concatenation of spatial features with global features is used as a feature map for the classification stage in the committee network. Two committee networks are fed separately with LBP and raw images. After training the two committee networks, to classify the emotion, the maximum probability between the two networks is considered as the final output. The proposed method was applied and tested on the FER2013 dataset. Our proposed method is more accurate than many leading methods, and in competition with the successful model that has a more complex architecture and higher computational cost, it has been able to achieve acceptable results with a simple architecture.
{"title":"Facial Expression Recognition using Spatial Feature Extraction and Ensemble Deep Networks","authors":"E. Afshar, Hassan Khotanlou, Elham Alighardash","doi":"10.1109/IPRIA59240.2023.10147196","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147196","url":null,"abstract":"Researchers have shown that 55% of concepts are conveyed through facial emotion and only 7% are conveyed by words and sentences, so facial expression plays an important role in conveying concepts in human communications. In recent years, due to the improvement of artificial neural networks, many studies have been conducted related to facial expression recognition. This paper presents a method based on ensemble classification using convolutional neural networks to recognize facial emotions. The concatenation of spatial features with global features is used as a feature map for the classification stage in the committee network. Two committee networks are fed separately with LBP and raw images. After training the two committee networks, to classify the emotion, the maximum probability between the two networks is considered as the final output. The proposed method was applied and tested on the FER2013 dataset. Our proposed method is more accurate than many leading methods, and in competition with the successful model that has a more complex architecture and higher computational cost, it has been able to achieve acceptable results with a simple architecture.","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132124392","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}
Pub Date : 2021-10-07DOI: 10.1109/IPRIA59240.2023.10147168
Atrin Arya, Hanieh Naderi, S. Kasaei
Recent research has revealed that the security of deep neural networks that directly process 3D point clouds to classify objects can be threatened by adversarial samples. Al-though existing adversarial attack methods achieve high success rates, they do not restrict the point modifications enough to preserve the point cloud appearance. To overcome this shortcoming, two constraints are proposed. These include applying hard boundary constraints on the number of modified points and on the point perturbation norms. Due to the restrictive nature of the problem, the search space contains many local maxima. The proposed method addresses this issue by using a high step-size at the beginning of the algorithm to search the main surface of the point cloud fast and effectively. Then, in order to converge to the desired output, the step-size is gradually decreased. To evaluate the performance of the proposed method, it is run on the ModelNet40 and ScanObjectNN datasets by employing the state-of-the-art point cloud classification models; including PointNet, PointNet++, and DGCNN. The obtained results show that it can perform successful attacks and achieve state-of-the-art results by only a limited number of point modifications while preserving the appearance of the point cloud. Moreover, due to the effective search algorithm, it can perform successful attacks in just a few steps. Additionally, the proposed step-size scheduling algorithm shows an improvement of up to 14.5% when adopted by other methods as well. The proposed method also performs effectively against popular defense methods.
{"title":"Adversarial Attack by Limited Point Cloud Surface Modifications","authors":"Atrin Arya, Hanieh Naderi, S. Kasaei","doi":"10.1109/IPRIA59240.2023.10147168","DOIUrl":"https://doi.org/10.1109/IPRIA59240.2023.10147168","url":null,"abstract":"Recent research has revealed that the security of deep neural networks that directly process 3D point clouds to classify objects can be threatened by adversarial samples. Al-though existing adversarial attack methods achieve high success rates, they do not restrict the point modifications enough to preserve the point cloud appearance. To overcome this shortcoming, two constraints are proposed. These include applying hard boundary constraints on the number of modified points and on the point perturbation norms. Due to the restrictive nature of the problem, the search space contains many local maxima. The proposed method addresses this issue by using a high step-size at the beginning of the algorithm to search the main surface of the point cloud fast and effectively. Then, in order to converge to the desired output, the step-size is gradually decreased. To evaluate the performance of the proposed method, it is run on the ModelNet40 and ScanObjectNN datasets by employing the state-of-the-art point cloud classification models; including PointNet, PointNet++, and DGCNN. The obtained results show that it can perform successful attacks and achieve state-of-the-art results by only a limited number of point modifications while preserving the appearance of the point cloud. Moreover, due to the effective search algorithm, it can perform successful attacks in just a few steps. Additionally, the proposed step-size scheduling algorithm shows an improvement of up to 14.5% when adopted by other methods as well. The proposed method also performs effectively against popular defense methods.","PeriodicalId":109390,"journal":{"name":"2023 6th International Conference on Pattern Recognition and Image Analysis (IPRIA)","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130178807","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：