نشریه ماشین بینایی و پردازش تصویر
سال نهم شماره 2 (تابستان 1401)

The kinship Verification system analyzes the facial features of two people to determine whether they are related or not. To identify the kinship, different features can be extracted from the faces. In this paper, to evaluate a kinship verification system for the first-generation kinship (father-son, father-daughter, mother-son, and mother-daughter), texture and color features are tested, and feature fusion, as well as examining several different classifiers is considered. In this regard, two proposed approaches have been proposed: (1) fusing effective features and evaluate different classifiers for kinship verification and (2) using NRML metric learning to generate a distinctive feature vector to increase kinship verification efficiency. The proposed methods for the two databases KinFaceW-I and KinFaceW-II have been analyzed and evaluated in different cases. The results of the evaluations show that the fusion of features and the use of NRML metric learning have been able to improve the performance of the kinship verification system. In addition to the two proposed approaches, feature extraction from the whole image as well as image blocks is proposed and the results are presented. The results indicate that using the block-wise method for feature extraction can be effective in improving the final kinship verification results.

The purpose of design and building autonomous cars is to eliminate the human factor in order to reduce losses and costs and also increase safety by replacing smart equipment. Todays, using artificial intelligence and machine learning, we are witnessing significant advances in the intelligent transportations, especially fully automated vehicles, which are able to analyze environmental information using advanced sensors and machine vision techniques. One of the challenges in designing such systems is a correct identification of other vehicles around the route of the vehicle. In this paper, a deep learning-based method for identifying emergency vehicles is presented in which feature extraction and classification processes are performed simultaneously. The deep network used in this research is a convolutional network. In Convolutional Neural Networks (CNN), achieving acceptable results and proper performance requires having a huge amount of data for network training. Due to the limited number of images in the data set used in this study and in order to increase the identification accuracy, transfer learning process and VGG16 pre-trained network have been used. Two new datasets were created for this study and furthermore two other known datasets were also examined. The proposed method was compared with four other known methods from the literature, where the final results showed supremacy of the proposed approach.

Gait disorder is an issue in human society that can limit a person's life activity as an injury. Correct and rapid diagnosis of this disorder can help the healing process. Therefore, research has always been done to identify this disorder, based on which the physician can consider an appropriate treatment plan for the patient.Many recent studies have used different wearable and non-wearable sensors for identification. In the present study, Kinect sensor, which can identify 25 points of anatomical points of a person's body, was used to receive and record data, the identification method is defined based on the Fuzzy System defined by a medical team consisting of specialists and knee surgeons as experts in this system. The number of participants is 90, which is given from the data of 45 people in the training step and construction of the fuzzy system based on the opinions of experts, and in the second step from the data of another 45 people to evaluate the system as input. The outputs of the fuzzy system validation show that based on 7 defined features and rules made in the fuzzy system, the accuracy, precision, and recall are reported are 95%, 92.14%, and 89.26%, respectively.

Local binary pattern is one of the most popular descriptor that widely used in feature extraction of texture images. Deep convolutional neural network is also one of the best classification methodthat provides very high accuracy. In this research, by combining the features that produced by these two methods, a structure for noisy texture classification is proposed, which provides a very high classification rate. This method is based on two extracted features. The first part uses completed local binary pattern features and in the second part the features of texture images are extracted by using the DenseNet-121 convolution deep neural network. Another motivation of this research related to feature reduction, which significantly reduces the dimensions of extracted features. It employs a shallow convolution neural network to convert the extracted features into lower number of new features. The accuracy of the proposed method has been evaluated on noisy Outex, CUReT and UIUC datasets. The classification accuracy of the proposed method for different level of noise has increased significantly compared to many advanced methods and has improved between 3 and 25%.

One of the issues in medical science, which has attracted the attention of many researchers, isliver segmentation from computer tomography images. Because the first step in the process of diagnosis of liver illnesses and its tumors is, having an appropriate image of the segmented liver in these images. The purpose of this paper is to provide an automated algorithm for liver segmentation in the CT images. Previous research has shown that the use of texture feature results in more favorable results in liver segmentation. The proposed algorithm of this paper is based on texture analysis to liver segmentation using the Kirsch edge detector, Mean shift, and k-means clustering. Results of the implementation of the proposed algorithm on 400 images of Milad hospital in Tehran containing liver and its lateral organs, showed the average of Dice criterion of 96%. Also, in the performance of the proposed algorithm on the sliver07 database, the average of Dice criterion is equal to 96.86%. Therefore, the proposed algorithm can be used as the first step in the process of diagnosis of liverillnesses and its tumors.

License Plate Detection is an essential part of smart transportation systems which is always ‎expected to be accurate and efficient. The exponential growth of the number of vehicles has led ‎to many problems for vehicle detection in different fields such as traffic control, parking lot, toll ‎highways and so. Using an automatic system for plate number detection of the vehicle, a large ‎number of problems can be solved. One of the best methods for overcoming this problem is using ‎a convolutional neural network approach. They have shown impressive performance in various ‎computer vision tasks. In this work, we tried to introduce a new dataset of Iranian cars' plates. ‎Then using transfer learning and Alexnet, we have proposed a new approach for detection and ‎classifying the plate numbers. We conducted comprehensive experiments on the new dataset. The ‎experimental results show that the proposed method has achieved to 97.35% accuracy.‎