Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
Shot classification for human behavioural analysis in video surveillance applications
No full textHuman behavior analysis plays a vital role in ensuring security and safety of people in crowded public places against diverse contexts like theft detection, violence prevention, explosion anticipation etc. Analysing human behaviour by classifying of videos in to different shot types helps in extracting appropriate behavioural cues. Shots indicates the subject size within the frame and the basic camera shots include: the close-up, medium shot, and the long shot. If the video is categorised as Close-up shot type, investigating emotional displays helps in identifying criminal suspects by analysing the signs of aggressiveness and nervousness to prevent illegal acts. Mid shot can be used for analysing nonverbal communication like clothing, facial expressions, gestures and personal space. For long shot type, behavioural analysis is by extracting the cues from gait and atomic action displayed by the person. Here, the framework for shot scale analysis for video surveillance applications is by using Face pixel percentage and deep learning based method. Face Pixel ratio corresponds to the percentage of region occupied by the face region in a frame. The Face pixel Ratio is thresholded with predefined threshold values and grouped into Close-up shot, mid shot and long shot categories. Shot scale analysis based on transfer learning utilizes effective pre-trained models that includes AlexNet, VGG Net, GoogLeNet and ResNet. From experimentation, it is observed that, among the pre-trained models used for experimentation GoogLeNet tops with the accuracy of 94.61%
Deep Learning Based Localisation and Segmentation of Prostate Cancer from mp-MRI Images
No full textProstate Cancer (PCa) is one of the most common diseases in adult males. Currently, mp-MRI imaging represents the most promising technique for screening, diagnosing, and managing this cancer. However, the multiple mp-MRI sequences\u27 visual interpretation is not straightforward and may present crucial inter-reader variability in the diagnosis, especially when the images contradict each other. In this work, we propose a computer-aided diagnostic system to assist the radiologist inlocating and segmenting prostate lesions. As fully convolutional neural networks (UNet) have proved themselves the leading algorithm for biomedical image segmentation, we investigate their use to find PCa lesions and segment for accurate lesions contours jointly. We offer a fully automatic system via MultiResUNet, initially proposed to segment skin cancer. We trained and validated an altered version of the MultiResUnet model using an augmented Radboudumc prostate cancer dataset and obtained encouraging results. An accuracy of 98.34\% is achieved, outperforming the concurrent system based on deep architecture
Enhanced SVM Based Covid 19 Detection System Using Efficient Transfer Learning Algorithms
No full textThe detection of the novel coronavirus disease (COVID-19) has recently become a critical task for medical diagnosis. Knowing that deep Learning is an advanced area of machine learning that has gained much of interest, especially convolutional neural network. It has been widely used in a variety of applications. Since it has been proved that transfer learning is effective for the medical classification tasks, in this study; COVID -19 detection system is implemented as a quick alternative, accurate and reliable diagnosis option to detect COVID-19 disease. Three pre-trained convolutional neural network based models (ResNet50, VGG19, AlexNet) have been proposed for this system. Based on the obtained performance results, the pre-trained models with support vector machine (SVM) provide the best classification performance compared to the used models individually
Color Image Visual Secret Sharing with Expressive Shares using Color to Gray & Back and Cosine Transform
No full textColor Visual Secret Sharing (VSS) is an essential form of VSS. It is so because nowadays, most people like to share visual data as a color image. There are color VSS schemes capable of dealing with halftone color images or color images with selected colors, and some dealing with natural color images, which generate low quality of recovered secret. The proposed scheme deals with a color image in the RGB domain and generates gray shares for color images using color to gray and back through compression. These shares are encrypted into an innocent-looking gray cover image using a Discrete Cosine Transform (DCT) to make meaningful shares. Reconstruct a high-quality color image through the gray shares extracted from an innocent-looking gray cover image. Thus, using lower bandwidth for transmission and less storage
Feature selection based on discriminative power under uncertainty for computer vision applications
No full textFeature selection is a prolific research field, which has been widely studied in the last decades and has been successfully applied to numerous computer vision systems. It mainly aims to reduce the dimensionality and thus the system complexity. Features have not the same importance within the different classes. Some of them perform for class representation while others perform for class separation. In this paper, a new feature selection method based on discriminative power is proposed to select the relevant features under an uncertain framework, where the uncertainty is expressed through a possibility distribution. In an uncertain context, our method shows its ability to select features that can represent and discriminate between classes
Retinal Blood Vessels Segmentation using Fréchet PDF and MSMO Method
No full textBlood vessels of retina contain information about many severe diseases like glaucoma, hypertension, obesity, diabetes etc. Health professionals use this information to detect and diagnose these diseases. Therefore, it is necessary to segment retinal blood vessels. Quality of retinal image directly affects the accuracy of segmentation. Therefore, quality of image must be as good as possible. Many researchers have proposed various methods to segment retinal blood vessels. Most of the researchers have focused only on segmentation process and paid less attention on pre processing of image even though pre processing plays vital role in segmentation. The proposed method introduces a novel method called multi-scale switching morphological (MSMO) for pre processing and Fréchet match filter for retinal vessel segmentation. We have experimentally tested and verified the proposed method on DRIVE, STARE and HRF data sets. Obtained outcome demonstrate that performance of the proposed method has improved substantially. The cause of improved performance is the better pre processing and segmentation methods
A multiple secret image embedding in dynamic ROI keypoints based on hybrid Speeded Up Scale Invariant Robust Features (h-SUSIRF) algorithm
No full textThis paper presents a robust and high-capacity video steganography framework using a hybrid Speeded Up Scale Invariant Robust Features (h-SUSIRF) keypoints detection algorithm. There are two main objectives in this method: (1) determining the dynamic Region of Interest (ROI) keypoints in video scenes and (2) embedding the appropriate secret data into the identified regions. In this work, the h-SUSIRF keypoints detection scheme is proposed to find keypoints within the scenes. These identified keypoints are dilated to form the dynamic ROI keypoints. Finally, the secret images are embedded into the dynamic ROI keypoints’ locations of the scenes using the substitution method. The performance of the proposed method (PM) is evaluated using standard metrics Structural Similarity Index Measure (SSIM), Capacity (Cp), and Bit Error Rate (BER). The standard of the video is ensured by Video Quality Measure (VQM). To examine the efficacy of the PM some recent steganalysis schemes are applied to calculate the detection ratio and the Receiver Operating Characteristics (ROC) curve is analyzed. From the experimental analysis, it is deduced that the PM surpasses the contemporary methods by achieving significant results in terms of imperceptibility, capacity, robustness with lower computational complexity
Material Classification with a Transfer Learning based Deep Model on an imbalanced Dataset using an epochal Deming-Cycle-Methodology
No full textThis work demonstrates that a transfer learning-based deep learning model can perform unambiguous classification based on microscopic images of material surfaces with a high degree of accuracy. A transfer learning-enhanced deep learning model was successfully used in combination with an innovative approach for eliminating noisy data based on automatic selection using pixel sum values, which was refined over different epochs to develop and evaluate an effective model for classifying microscopy images. The deep learning model evaluated achieved 91.54% accuracy with the dataset used and set new standards with the method applied. In addition, care was taken to achieve a balance between accuracy and robustness with respect to the model. Based on this scientific report, a means of identifying microscopy images could evolve to support material identification, suggesting a potential application in the domain of materials science and engineering. 
A neural network with competitive layers for character recognition
No full textA structure and functioning mechanisms of a neural network with competitive layers are described. The network is intended to solve the character recognition task. The network consists of several competitive layers of neurons. Each layer is a neural network consisting of a number of neurons represented as a layer. The number of neural layers is equal to the number of recognized classes. All neural layers have one-to-one correspondence with one another and with the input raster. The neurons of every layer have mutual lateral learning connections, which weights are modified during the learning process. There is a competitive (inhibitory) relationship between all neural layers. This competitive interaction is realized by means of a “winner-take-all” (WTA) procedure which aim is to select the layer with the highest level of neural activity.Validation of the network has been done in experiments on recognition of handwritten digits of the MNIST database. The experiments have demonstrated that its error rate is few less than 2%, which is not a high result, but it is compensated by rather fast data processing and a very simple structure and functioning mechanisms.
Attention-based CNN-ConvLSTM for Handwritten Arabic Word Extraction
No full textWord extraction is one of the most critical steps in handwritten recognition systems. It is challenging for many reasons, such as the variability of handwritten writing styles, touching and overlapping characters, skewness problems, diacritics, ascenders, and descenders\u27 presence. In this work, we propose a deep-learning-based approach for handwritten Arabic word extraction. We used an Attention-based CNN-ConvLSTM (Convolutional Long Short-term Memory) followed by a CTC (Connectionist Temporal Classification) function. Firstly, the text-line input image\u27s essential features are extracted using Attention-based Convolutional Neural Networks (CNN). The extracted features and the text line\u27s transcription are then passed to a ConvLSTM to learn a mapping between them. Finally, we used a CTC to learn the alignment between text-line images and their transcription automatically. We tested the proposed model on a complex dataset known as KFUPM Handwritten Arabic Text (KHATT \cite{khatt}). It consists of complex patterns of handwritten Arabic text-lines. The experimental results show an apparent efficiency of the used combination, where we ended up with an extraction success rate of 91.7\%