sajad mohamadzadeh

The expansion of touch-screen devices has provided the possibility of human-machine interactions in the form of free-hand drawings. In sketch-based image retrieval (SBIR) systems, the query image is a simple binary design that represents the mental image of a person with the rough shape of an object. A simple sketch is convenient and efficient for recording ideas visually, and can outdo hundreds of words. The objective is to retrieve a natural image with the same label as the query sketch. This article presents a multi-step training method. Regression functions are used in the deep network structure to improve system performance, and various loss functions are employed for a better convergence of the retrieval system. The convolutional neural network used has two branches, one related to the sketch and the other related to the image, and these two branches can have the same or different architecture. After four training steps, a 56.48% MAP was achieved, indicating the desirable performance of the network.

This study proposes a descriptor-based approach combined with deep learning, which recognizes facial emotions for safe driving.  Paying attention to the driver's facial expressions is crucial to address the increasing road accidents. This project aims to develop a Facial Emotion Recognition (FER) system that monitors the driver's facial expressions to identify emotions and provide instant assistance for safety control. In the initial stage, Viola-Jones face detection was employed to detect the facial region, followed by Butterworth high-pass filtering to enhance the identified region for locating the eye, nose, and mouth regions, using Viola-Jones face detection. Secondly, the Local Binary Patterns (LBP) feature descriptor is utilized to extract features from the identified eye, nose, and mouth regions. Using 3 RGB channels, the extracted features from these three regions are fed into RessNet-50 and EfficientNet deep networks. The outputs of the two deep learning models' classifiers are combined and integrated using two ensemble methods ensemble maximum voting and ensemble mean. Based on these combining classifier rules, the performance was evaluated on the JAFFE and KMU-FED databases. The experimental results demonstrate that the proposed method can effectively and with higher accuracy than other competitors recognize emotions in the JAFFE and KMU-FED datasets. The novelty and originality of this paper lie in its significant application in the automotive industry. Implementing our proposed method in a system capable of high accuracy and precision can help mitigate numerous driving hazards. Our approach has achieved 99% and 98% accuracy on the JAFFE and KMU-FED databases, respectively. This high level of accuracy, coupled with its practical relevance, underscores the innovative nature of our work.

Prostate cancer is considered as one of the most common diseases of men worldwide after lung cancer in this century. Diagnosis and classification of medical images is a very challenging task. Timely diagnosis and prevention of this disease helps significantly in the treatment process and preventing its progression to other tissues. In order to determine the degree of the disease, the tissues are sampled and the type of degree is determined by examining the pathology structure. The most common method used to diagnose prostate cancer is microscopic observation of stained biopsies by a pathologist and Gleason grading of pathology image. In the most recent classification, the prostate tissue is divided into five grades, where grade one is the most benign condition and grade five indicates the worst condition of the disease. In this article, a new method has been introduced that combines the output of the classifications of three models of deep learning. The Transfer Learning method trained on a set of images (ImageNet) has been used. To evaluate the trained networks, test samples are applied to the networks. By using two rules of the combination of classifiers, such as majority voting and averaging, the test image will be assigned to a class. In order to evaluate the proposed method, this method has been applied to the database of the International Society of Urological Pathology (ISUP). The proposed method in this research is based on ensemble deep learning, which provides detection accuracy about 97.2% and 97.33% in prostate cancer according, respectively, for two rules of combining voting and averaging. Compared to other conventional methods for grading prostate cancer pathology images, it has increased the accuracy of more than 4% in diagnosis and classification.

Nowadays, one of the most principal processes of perceiving the content of videos is the moving object tracking, in which the tracking process of a moving object is implemented in each video frame. The use of filters in this field has been increased during the last decade. Correlation filters are one of the most widely used filters in this field, however, using this filter, as usual, may cause the problem of target drifting. The present study proposes a novel method to improve the performance of the correlation filter. The advanced searching strategy can greatly overcome to the problem of target drifting with examining a threshold level by calculating the average and variance in each frame. In this regard, if the level of the threshold is reduced, a mechanism will be activated to recover the target in the current frame. In order to describe the target, the histogram of the oriented gradients feature has been used because this feature shows the changes in illumination variation better than other features. The proposed method has been examined in single-camera mode on the TB50 and TB100 datasets. To evaluate the proposed method, several criteria including precision, correct detection rate (CDR), average location error (ALE), and frame per second are used. The obtained results on the TB50 show that the proposed method, compared to the KCF method, has achieved an improvement around 9% in the precision criterion, improvement 6% in the success rate criterion and a 50% reduction in the ALE criterion. Also, the obtained results on TB100 have increased the precision criterion by approximately 15%, the success rate by 12%, and a favorable reduction of 50% on the ALE.

Pathological analysis plays an important role in the diagnosis, prediction and planning of cancer treatment. Using digital pathology, ie, scanning and storing digital parts of patient tissue, tools for analyzing these complex images now can be developed. Doctors use a computer diagnostic system from an intelligent assistant to accurately diagnose. These systems have great benefits in improving treatment efficacy.

In this study, the deep neural network classifier has been used with the help of the Tensor Flow framework and the use of the Keras-library. Input images are initially transmitted from a low pass filter to reduce noise effects. The pre-processed images are then imported into a convolutional neural network.

The results of the research reveal a significant difference in the accuracy values between different methods with the proposed method, which in some cases indicates an increase of more than 14.18% in the accuracy of the diagnosis. Another advantage of the proposed method is to provide high sensitivity to histopathologic images, which shows an increase of 12 to 18 percent compared to other studies. The reason for this is the excellence of extracting high-level features through convolutional neural network, which is accompanied by a reduction in the size of the feature vector.

The results showed a accuracy of %98.6 for skin lesions and %96.1 accuracy for breast cancer histopathologic findings, which offers promising results compared to the results of other studies.

Salient object detection attracted the attention of researchers in various fields, so that it is used in many applications of the visual machine, such as object detection and tracking. Most of the work in this area is based on bottom-top saliency models and use low-level features to extract the final saliency map that these works do not have a significant accuracy in salient object detection. On the other hand, top-down visual models are used for the specific applications. In this paper, a different method has been proposed to extract the salient object map that uses low-up and top-down attributes for extracting the salient and it is based on the learning process. The simultaneous selection of these features makes the proposed algorithm for various applications and increases the accuracy of the salient object detection. The learning process is performed by using the and Convolutional Neural Network (CNN) structure. After the decomposition of image to its super pixels, different features of image are extracted. Then, the extracted features are normalized to have zero mean and unit variance, and CNN is used to train the features. The accuracy of the proposed method is improved by using of the 8 salient and CNN. The performance of the proposed method has been compared to twenty method by applying four new databases including MSRA-100, ECSSD, MSRA-10K and Paskal-S. The proposed method provides better results compared to the other methods in respect to prediction of salient object.

The importance and demand of visual scene understanding have been increasing because of autonomous systems development. Optical flow is known as an important tool for scene understanding. Current optical flow methods present general assumptions and spatial homogeneous for spatial structure of flow. In fact, the optical flow in an image depends on object class and the type of object movement. The first assumption in many methods in this field is the brightness constancy during movements of pixels between frames. This assumption is proven to be inaccurate in general. In this paper, we use recent development of deep convolutional networks in semantic segmentation of static scenes to divide an image in to different objects and also depends on type of the object different movement patterns are defined. Next, estimation of the optical flow is performed by using deep neural network for initial image which has been semantically segmented. The proposed method provides minimum error in optical flow measures for KITTI-2015 database and results in more accurate segmentation compared to state-of-the-art methods for several natural videos.

Detecting social groups is one of important and complex problems which has been concerned recently. Detecting social groups and relation between group members will be necessary for human robots in near future. Databases have some information including trajectories and also labels of members. The target is to detect social groups that contains at least two people or detecting individual motion of the persons. In the proposed method, for detecting social groups, physical distance, temporal causality and shape similarity features are used. The required time to extract these features is lower than the other suggested features. In addition to accuracy, the effectiveness of the proposed method in terms of required time for training and testing data is also examined. Lower required time provides greater ability to implement for human robots. The proposed method provides acceptable results in valid databases and is compared to existing methods in terms of statistical results and the required time.

Nowadays, with huge progress in digital imaging, new image processing methods are needed to manage digital images stored on disks. Image retrieval has been one of the most challengeable fields in digital image processing which means searching in a big database in order to represent similar images to the query image. Although many efficient researches have been performed for this topic so far, there is a semantic gap between human concept and features extracted from the images and it has become an important problem which decreases retrieval precision. In this paper, a convolutional neural network (CNN) is used to extract deep and high-level features from the images. Next, an optimization problem is defined in order to model the retrieval system. Heuristic algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO) have shown an effective role in solving the complex problems. A recent introduced heuristic algorithm is Grasshopper Optimization Algorithm (GOA) which has been proved to be able to solve difficult optimization problems. So, a new search method, modified grasshopper optimization algorithm (MGOA) is proposed to solve modeled problem and to retrieve similar images efficiently, despite of total search in database. Experimental results showed that the proposed system named CNN-MGOA achieves superior accuracy compared to traditional methods.

Recent researches on pixel-wise semantic segmentation, use deep neural networks to improve accuracy and speed of these networks in order to increase the efficiency in practical applications such as automatic driving. These approaches have used deep architecture to predict pixel tags, but the results seem to be undesirable. The reason for these unacceptable results is mainly due to the existence of max pooling operators, which reduces the resolution of the feature maps. In this paper, we present a convolutional neural network composed of encoder-decoder segments based on successful SegNet network. The encoder section has a depth of 2, which in the first part has 5 convolutional layers, in which each layer has 64 filters with dimensions of 3×3. In the decoding section, the dimensions of the decoding filters are adjusted according to the convolutions used at each step of the encoding. So, at each step, 64 filters with the size of 3×3 are used for coding, the weights of these filters are adjusted by network training and adapted to the educational data. Due to having the low depth of 2, and the low number of parameters in the proposed network, the speed and the accuracy improve in comparison with the popular networks such as SegNet and DeepLab. For the CamVid dataset, after a total of 60,000 iterations, we attain the 91% for global accuracy, which indicates improvements in the efficiency of the proposed method

Skin cancer has grown dramatically over the past decades, and the importance of early treatment is increasing day by day. The purpose of this study is to use deep neural networks to create an auto-diagnosis system for melanoma, in which data is directly controlled as part of a deep learning process. In this paper, studies on related pictures of skin cancer were performed.  For the diagnosis of benign or malignant skin cancer, the deep neural network classifier is used with the help of the Tensorflow framework and the use of the Keras libraries. The dataset which are used in this study consist 70 images of melanoma and 100 images of benign moles. In the proposed model, 80% of the database images are used for training and 20% of the database images are selected for testing. The proposed method offers a higher detection accuracy than other existing methods, which has increased the accuracy of diagnosis in most cases by more than 10%. The high accuracy of the diagnosis and classification and the speed of convergence to the final result are the characteristics of this Research Compared to other Research. An automatic system based on deep learning is presented to identify and categorize skin cancer which provides high accuracy and speed.

Abstract In recent years, interest in research into the application of intelligent algorithms for diagnosis and categorization of diseases, especially cancer has increased dramatically. Tumor classification is an important task in medical diagnosis. Technological calculations are important due to their classification function in diagnosis of medical illnesses. Diagnosing and classifying medical images is a challenging task. To detect the malignancy of prostate cancer and the opioid or malignant breast cancer, deep neural network classifier, which is based on Tensor flow framework and Keras library, is used. In the training phase, educational images are considered along with the output class for the network. During training, the weight of the filter is updated every time. However, after several replications, optimal weights are updated and the network is trained to extract the best feature from the images. In this research, the proposed method due to using deep neural network and accurate feature extraction provides detection accuracy about 95.83% and 99.5% for breast and prostate cancers, respectively, which is more than 7% compared to other methods. Cancer is one of the most prevalent diseases in the world. Cancer is started from the cells, which are the basic building blocks making the tissue. One of the challenges in medical diagnostic techniques is the difficulty in analyzing dense tissues. Since the detection of the diagnosis by human is time-consuming and has a higher probability of error, the researchers have been trying to detect it automatically by using different algorithms.

Traditionally, object retrieval methods require a set of images of a specific object for training. In this paper, we propose a new object retrieval method using a single query image, without training, for a global object. The query image could be a typical real image of the object. The object is constructed based on Speeded Up Robust Features (SURF) points acquired from the image. Information of relative positions, scale and orientation between SURF points are calculated and constructed into an object model. The ability to match partially affine transformed object images results from the robustness of SURF points and the flexibility of the model. Occlusion is handled by specifying the probability of a missing SURF point in the model. Experimental results show that this matching technique is robust under partial occlusion and rotation. The obtained results illustrate that the proposed method improves efficiency, speeds up recovery and reduces the storage space.

Image retrieval is one of the most applicable image processing techniques which have been used extensively. Feature extraction is one of the most important procedures used for interpretation and indexing images in content-based image retrieval (CBIR) systems. Reducing dimension of feature vector is one of challenges in CBIR systems. There are many proposed methods to overcome these challenges. However, the rate of image retrieval and speed of retrieval is still an interesting field of researches. In this paper we propose a new method based on combination of Hadamard matrix, discrete wavelet transform (HDWT2) and discrete cosine transform (DCT) and we used principal component analysis (PCA) to reduce dimension of feature vector and K-nearest neighbor (KNN) for similarity measurement. The precision at percent recall and ANR are considered as metrics to evaluate and compare different methods. Obtaining results show that the proposed method provides better performance in comparison with other methods.