The detection of vehicles and pedestrians on the road is one of the most challenging problems in object detection and autonomous vehicles. This paper reports a novel intelligent traffic monitoring and management system using You Only Look Once (YOLO) and OpenCV tracker. A new database named 'Kannur University Vehicle Database (KNUVDB)’ is created and used for the purpose of studying vehicle detection., in addition to the available datasets in literature. We focus on counting the vehicles after they have been detected and tracked. Later., performs the traffic update by using the vehicle count. The proposed method provides better detection accuracy on the real-time traffic video dataset available in the literatures and also on the KNUVDB dataset. Experimental studies have shown that Discriminative Correlation Filter with Channel and Spatial Reliability (CSRT) and Kernelized Correlation Filter (KCF) provide better performance than other OpenCV trackers. In KNUVDB dataset., YOLO-CSRT gives 100% accuracy and YOLO-KCF provides 90.90% accuracy in vehicle detection., tracking and counting. In real-time road traffic video dataset., YOLO-CSRT provides 100% accuracy and YOLO-KCF provides 93.70 % accuracy in vehicle detection., tracking., and counting accuracy.

Vehicle detection and tracking plays an important role in Intelligent Transportation Systems (ITS). This paper reports an improved vehicle detection and tracking performance of combined You Only Look Once (YOLO) and Discriminative Correlation Filter (with Channel and Spatial Reliability) (CSRT). CSRT is mainly used for face prediction and moving object detection. The proposed system uses CSRT for vehicle tracking, particularly for cars, buses, and trucks. To perform the vehicle detection task, we have used the YOLO v3 pre-trained model. The accuracy and effectiveness of our vehicle detection and tracking system are tested with 8 different commonly available trackers in various publicly available traffic videos. Experimental results show that the CSRT gives the best performance among all the other trackers. CSRT gives 100% accuracy in all the four publicly available traffic videos even though vehicles with poor lighting conditions.

Malayalam is an Indian language with a vast collection of traditional documents from heterogeneous sources like palm-leaf manuscripts containing valuable information on traditional medicine and culture, the majority of which are not available in digital formats. Development of Optical Character Recognition (OCR) systems is essential to digitize such documents for effective utilization and preservation. On account of degradation due to ageing and mishandling, noises such as stains, creases, shadows, uneven illumination, complex backgrounds, bleed-through ink, and faint text, historical documents are particularly difficult to process. The presence of such noises degrades the overall performance of the OCR system, if it is not preprocessed appropriately. This paper proposes a novel approach to denoise the document image in the preprocessing stage using division normalization and Otsu's thresholding to enhance its overall quality. The efficacy of the proposed approach is evaluated on document images with natural and synthetically introduced noise elements and it is observed that this work outperforms the pure Otsu thresholding-based method.

This paper presents a new real-time intelligent traffic monitoring system. To perform the vehicle detection, a filtered You Only Look Once (YOLO) is used. The pre-trained YOLO framework can detect 80 objects. The proposed system is tested for three classes of vehicles such as bus, truck, and car. After extracting the three categories, to obtain the count of that vehicle in each lane, checkpoint is assigned. The count is used to control the real-time road traffic signal. The system is tested with three different publicly available traffic videos. In the present work, we have used Kernel Correlation Filter (KCF) tracker and the object retrieval accuracy is obtained. Experimental results show that YOLO and KCF outperform Scale Invariant Feature Transform (SIFT) and Region-based Convolutional Neural Network (RCNN) with KCF tracker, and Maximally Stable Extremal Regions (MSER) and faster RCNN with KCF tracker.

Exudates detection is a main step in diabetic retinopathy diagnosis. Hard Exudates will be seen as yellow coloured deposits with clear borders. Automatic exudate detection is not possible yet as there is no good software available. In this paper, Hard Exudate detection which is the moderate stage detection of diabetic retinopathy is performed. Here, we use Deep Convolutional Neural Network as classifier and conduct experimental study by using DIARETDB1 database. Sensitivity of 100% and accuracy of 98.88% were obtained.

The paper analyzes the effectiveness of histogram-oriented gradient (HOG) features in static hand gesture recognition (SHGR) using k-nearest neighbor (k-NN) and support vector machine (SVM) classifiers. The experiment is carried out on a database of 10 different static gestures built on environmental conditions that are limited. A histogram-oriented gradient is extracted from the segmented gestures after performing various color image enhancements on the original database. For both classifiers k-NN and SVM, 80% of the extracted features are used for training and the remaining 20% are used for testing. For k-NN an average recognition accuracy of 96% is obtained. On the other hand, for SVM an average recognition accuracy of 98% is obtained on the same set of training and test data and found that the SVM classifier gives better performance than the k-NN classifier.

Object detection and retrieval are one of the fast-growing fields of computer vision. Retrieval of objects in images can be performed through either traditional approaches or deep learning-based approaches. A method known as Faster Region-based Convolutional Neural Network (R-CNN) was used for retrieving the same object in different scenes. Faster R-CNN was introduced to overcome the problems with the fast R-CNN framework. For enhancing feature extraction in the same object retrieval using faster R-CNN, we are imposing an existing algorithm called Maximally Stable Extremal Regions (MSER) which will enhance the features of low-resolution images. MSER can extract several regions from an image. The collaborative proposed model uses the prominent features of both faster R-CNN and MSER algorithm to retrieve a queried object from a given dataset with better accuracy. Experimental results have shown that our collaborative model achieved a retrieval efficacy in terms of mean Average Precision (mAP) 0.79 in the Vehicle dataset and 0.632 in the Oxford Building dataset. The Collaborative MSER and faster R-CNN model outperform with a mAP of 0.86 in the Vehicle dataset and 0.714 in the Oxford Building dataset.

Diabetic Retinopathy (DR) is a fast-growing retinal disease happens as a result of exponential growth in sugar level in blood which diminishes eyesight. The severity level identification of this eye disorder is performed by ophthalmologists due to scarcity of good software for finding DR. The initial stage of diabetic retinopathy is identified by the presence of microaneurysm. This paper conducts the initial phase detection of the disease by using Convolutional Neural Network (CNN). For conducting experiment DIARETDB1 dataset used. The images from the database are resized as a preprocessing step then automatic feature extraction done by the simple CNN used. By performing training, the CNN network classifies images with and without disease. The Sensitivity, Specificity and Accuracy obtained by the technique explained is 97.62%, 100% and 97.75%.

Object detection and retrieval is an active area of research. This paper proposes a collaborative approach that is based on multi-resolution maximally stable extreme regions (MRMSER) and faster region-based convolutional neural network (FRCNN) suitable for efficient object detection and retrieval of poor resolution images. The proposed method focuses on improving the retrieval accuracy of object detection and retrieval. The proposed collaborative model overcomes the problems in a faster RCNN model by making use of multi-resolution MSER. Two different datasets were used on the proposed system. A vehicle dataset contains three classes of vehicles and the Oxford building dataset with 11 different landmarks. The proposed MRMSER-FRCNN method gives a retrieval accuracy 84.48% on Oxford 5k building dataset and 92.66% on vehicle dataset. Experimental results show that the proposed collaborative approach outperform the faster RCNN model for poor-resolution conditioned query images.

The changes in complexity of brain while using two types of mobile phones are analyzed using Electroencephalogram (EEG) signals in this paper. Largest Lyapunov Exponent (LLE) of the data set prepared with EEG record of 35 subjects with and without phone is analyzed using non parametric Mann-Whitney U test. The test is significant in 8 electrodes for data set with phone-1 and is significant for one electrode for the data set with phone-2. Symmetry of brain using asymmetry index is analyzed for further verification and found within allowable range. It is concluded that there is no significant change in complexity of brain while using mobile phone.

The main intention of this research is to develop a novel ranking measure for content-based image retrieval system. Owing to the achievement of data retrieval, most commercial search engines still utilize a text-based search approach for image search by utilizing encompassing textual information. As the text information is, in some cases, noisy and even inaccessible, the drawback of such a recovery strategy is to the extent that it cannot depict the contents of images precisely, subsequently hampering the execution of image search. In order to improve the performance of image search, we propose in this work a novel algorithm for improving image search through a multi-kernel fuzzy c-means (MKFCM) algorithm. In the initial step of our method, images are retrieved using four-level discrete wavelet transform-based features and the MKFCM clustering algorithm. Next, the retrieved images are analyzed using fuzzy c-means clustering methods, and the rank of the results is adjusted according to the distance of a cluster from a query. To improve the ranking performance, we combine the retrieved result and ranking result. At last, we obtain the ranked retrieved images. In addition, we analyze the effects of different clustering methods. The effectiveness of the proposed methodology is analyzed with the help of precision, recall, and F-measures.

Diabetic Retinopathy (DR) one of the severe eye disorder which causes damage to capillaries in retina due to increase in blood sugar levels. Process which examines DR and its severity is presently performed by eye specialists due to the unavailability of good automated DR screening software. This paper proposes a method intended for early stage identification of DR with more accurate results compared to the existing methods. Key features extracted and classification using the extracted key features is performed using feed forward neural networks. For performing training and testing DIARETDB1 database is used. The Accuracy obtained is 98.89% for the proposed system.

To minimize the radiation risk in clinical examination using X-ray CT, low dose CT imaging is desirable in clinical practice. Lowering radiation dose degrades the reconstructed image quality and decreases the diagnostic performance of the image. Hence image denoising and signal enhancement in low dose CT imaging is one of the foremost challenging issue. The NLM algorithm has been successfully applied in removing noise and artefacts from corrupted natural images. This paper studies the application of non-local means (NLM) algorithm in removing noise from low dose CT lung images. In this study low dose CT images are denoised by applying NLM filer in two different ways. In the first application NLM filter is used to remove noise from projection data before reconstructing the images. In the second method images are denoised by applying NLM in the image domain after reconstruction. The Performances of both image space and projection space denoising is compared and evaluated quantitatively using PSNR and SSIM. Results shows that NLM filter can successfully applied to denoise low dose CT images and better results can obtained if the filter is applied on reconstructed images.

Image-based object retrieval has numerous applications in the field of machine vision to inquire from an appropriate image or video sequence for a given query object. The object retrieval task is conventionally carried out by a set of handcrafted algorithms, which provides image depictions in the fashion of visual characteristics. During the last decade, an extensive change has been practiced to describe visual content from handcrafted characteristics to the application of machine learning approaches and to the real layout of the image descriptors. The extensive movement is based on Convolutional Neural Networks (CNN) which is popularly known as Deep Learning. This proposed work deals with a combination of both conventional and machine learning approaches to retrieve an image object from a given dataset. This is done by a series of activities such as feature extraction and storage of training images, query image selection and feature extraction. Similarity matching between database and query image features. Final retrieval is based on the objectness score.

Edge detection is a crucial step in various image processing systems like computer vision , pattern recognition and feature extraction.The Canny edge detection algorithm even though exhibits high accuracy, is computationally more complex compared to other edge detection techniques.A block based distributed edge detection technique is presented in this paper, which adaptively finds the thresholds for edge detection depending on block type and the distribution of gradients in each block.A novel method of computation of high threshold has been proposed in this paper.Block-based hysteresis thresholds are computed using a non uniform gradient magnitude histogram.The algorithm exhibits remarkably high edge detection accuracy, scalability and significantly reduced computational time.Pratt's Figure of Merit quantifies the accuracy of the edge detector, which showed better values than that of original Canny and distributed Canny edge detector for benchmark dataset.The method detected all visually prominent edges for diverse block size.

Edge detection is a crucial step in various image processing systems like computer vision , pattern recognition and feature extraction. The Canny edge detection algorithm even though exhibits high accuracy, is computationally more complex compared to other edge detection techniques. A block based distributed edge detection technique is presented in this paper, which adaptively finds the thresholds for edge detection depending on block type and the distribution of gradients in each block. A novel method of computation of high threshold has been proposed in this paper. Block-based hysteresis thresholds are computed using a non uniform gradient magnitude histogram. The algorithm exhibits remarkably high edge detection accuracy, scalability and significantly reduced computational time. Pratt’s Figure of Merit quantifies the accuracy of the edge detector, which showed better values than that of original Canny and distributed Canny edge detector for benchmark dataset. The method detected all visually prominent edges for diverse block size.

Removal of Mixed noise from digital color images is a challenging task because it requires processing of different types of noise. Also, noise need to be distinguished from the original image structures such as edges and details. Fuzzy theory is an effective solution to this problem. In this paper, a new fuzzy method is proposed to reduce impulse and Gaussian noise from color images. A weighted averaging filtering operation is used for this purpose. The weights in the averaging process are assigned using a fuzzy rule system based on a new certainty function, so as to reduce both noise types and to preserve image structures. Experimental results show that the method outperforms the state-of-the-art filters.

Removing mixed noise from digital images is a challenging problem as it involves processing of various types of noise. Additionally, noises should be differentiated from inherent image structures. Fuzzy logic is a successful solution in this situation. In this paper, we propose a novel fuzzy technique to reduce mixed ImpulseGaussian noise from color images. A weighted averaging process is used in this regard. The weights are computed using a fuzzy rule system which operates on an improved certainty function. This method performs effectively in reducing both the noise types and in preserving image details. The simulation experiments demonstrated that the method outperforms state-of-the-art filter.