Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
Diabetic foot ulcer segmentation using logistic regression, DBSCAN clustering and mathematical morphology operators
No full textDigital images are used for evaluation and diagnosis of a diabetic foot ulcer. Selecting the wound region (segmentation) in an image is a preliminary step for subsequent analysis. Most of the time, manual segmentation isn\u27t very reliable because specialists could have different opinions over the ulcer border. This fact encourages researchers to find and test different automatic segmentation techniques. This paper presents a computer-aided ulcer region segmentation algorithm for diabetic foot images. The proposed algorithm has two stages: ulcer region segmentation, and post-processing of segmentation results. For the first stage, a trained machine learning model was selected to classify pixels inside the ulcer\u27s region, after a comparison of five learning models. Exhaustive experiments have been performed with our own annotated dataset from images of Cuban patients. The second stage is needed because of the presence of some misclassified pixels. To solve this, we applied the DBSCAN clustering algorithm, together with dilation, and closing morphological operators. The best-trained model after the post-processing stage was the logistic regressor (Jaccard Index , accuracy , recall , precision , and F1 score ). The trained model was sensitive to irrelevant objects in the scene, but the patient foot. Physicians found these results promising to measure the lesion area and to follow-up the ulcer healing process over treatments, reducing errors
Adaptive Window Selection for Non-uniform Lighting Image Thresholding
No full textSelection of appropriate size of windows or subimages is the most important step for thresholding images with non-uniform lighting. In this paper, a novel criteria function is developed to partition images into different size of sub images appropriate for thresholding. After the partitioning, each subimage is segmented by Otsu’s thresholding approaches. The performance of the proposed method is validated on benchmark test images with different degree of uneven lighting. Based on the qualitative and quantitative measures, the proposed method is fully automatic, fast and efficient in comparison to many landmark approaches
An Efficient BoF Representation for Object Classification
No full textThe Bag-of-features (BoF) approach has proved to yield better performance in a patch-based object classification system owing to its simplicity. However, often the very large number of patch-based descriptors (such as scale-invariant feature transform and speeded up robust features, extracted from images to create a BoF vector) leads to huge computational cost and an increased storage requirement. This paper demonstrates a two-staged approach to creating a discriminative and compact BoF representation for object classification. As a preprocessing stage to the codebook construction, ambiguous patch-based descriptors are eliminated using an entropy-based and one-pass feature selection approach, to retain high-quality descriptors. As a post-processing stage to the codebook construction, a subset of codewords which is not activated enough in images are eliminated from the initially constructed codebook based on statistical measures. Finally, each patch-based descriptor of an image is assigned to the closest codeword to create a histogram representation. One-versus-all support vector machine is applied to classify the histogram representation. The proposed methods are evaluated on benchmark image datasets. Testing results show that the proposed methods enables the codebook to be more discriminative and compact in moderate sized visual object classification tasks
Edge detection algorithm for omnidirectional images, based on superposition laws on Blach’s sphere and quantum entropy
No full textThis paper presents an edge detection algorithm for omnidirectional images based on superposition law on Bloch’s sphere and quantum local entropy. Omnidirectional vision system has become an essential tool in computer vision, duo to its large field of view. However, classical image processing algorithms are not suitable to be applied directly in this type of images without taking into account the spatial information around each pixel. To show the performance of the proposed method, a set of experimentation was done on synthetic and real images devoted to agriculture applications. Later, Fram & Deutsh criterion has been adopted to evaluate its performance against three algorithms proposed on the literature and developed for omnidirectional images. The results show a good performance of the proposed method in term of edge quality, edge community and sensibility to noise
A comparison of an RGB-D cameras performance and a stereo camera in relation to object recognition and spatial position determination
No full textResults of using an RGB-D camera (Kinect sensor) and a stereo camera, separately, in order to determine the 3D real position of characteristic points of a predetermined object in a scene are presented. KAZE algorithm was used to make the recognition, that algorithm exploits the nonlinear scale space through nonlinear diffusion filtering; 3D coordinates of the centroid of a predetermined object were calculated employing the camera calibration information and the depth parameter provided by a Kinect sensor and a stereo camera. Experimental results show it is possible to get the required coordinates with both cameras in order to locate a robot, although a balance in the distance where the sensor is placed must be guaranteed: no fewer than 0.8 m from the object to guarantee the real depth information, it is due to Kinect operating range; 0.5 m to stereo camera, but it must not be 1 m away to have a suitable rate of object recognition, besides, Kinect sensor has more precision with distance measures regarding a stereo camera
Underwater Acoustic Image Denoising Using Stationary Wavelet Transform and Various Shrinkage Functions
No full textUnderwater acoustic images are captured by sonar technology which uses sound as a source. The noise in the acoustic images may occur only during acquisition. These noises may be multiplicative in nature and cause serious effects on the images affecting their visual quality. Generally image denoising techniques that remove the noise from the images can use linear and non-linear filters. In this paper, wavelet based denoising method is used to reduce the noise from the images. The image is decomposed using Stationary Wavelet Transform (SWT) into low and high frequency components. The various shrinkage functions such as Visushrink and Sureshrink are used for selecting the threshold to remove the undesirable signals in the low frequency component. The high frequency components such as edges and corners are retained. Then the inverse SWT is used for reconstruction of denoised image by combining the modified low frequency components with the high frequency components. The performance measure Peak Signal to Noise Ratio (PSNR) is obtained for various wavelets such as Haar, Daubechies,Coiflet and by changing the thresholding methods
Saliency-Based Image Retrieval as a Refinement to Content-Based Image Retrieval
No full textSearching for an image in a database is important in different applications; hence, many algorithms have been proposed to identify the contents of the image. In some applications, but not all, identifying the content of the image as a whole can offer good results. Searching for an object inside the image is more important in most applications than identifying the image as a whole. Therefore, studies focused on segmenting the image into small sub-images and identified their contents. In view of the concepts of human attention, various literature defined saliency as a computer representation of it, where different algorithms were developed to extract the salient regions. These salient regions, which are the regions that attract human attention, are used to identify the most important regions that contain important objects in the image. In this paper, we introduce a new algorithm that utilises the saliency principles to identify the contents of an image and search for similar objects in the images stored in a database. We also demonstrate that the use of salient objects produces better and more accurate results in the image retrieval process. A new retrieval algorithm is therefore presented here, focused on identifying the objects extracted from the salient regions. To assess the efficiency of the proposed algorithm, a new evaluation method is also proposed which considers the order of the retrieved image in assessing the efficiency of the algorithm
Processing historical photographs and film footage with Photogrammetry and Artificial Intelligence for Cultural Heritage documentation and virtual reconstruction
No full textThe specific objective of this thesis is to offer an excursion through the metric potentialities of different data available in historical archives, by considering the essential role of photogrammetry. The aim is to explore how metric information about buildings which no longer exist or transformed over time could be extracted from old photographs and videos of different quality, for their 3D virtual reconstruction analysing the material stored in historical archives to support researchers and experts in historical research of Cultural Heritage.In order to process these data and to obtain metrically certified results, a modification of the algorithms of the standard photogrammetric pipeline was necessary. This purpose was achieved with the use of open-source Structure-from-Motion algorithms and the creation of a specific benchmark to compare the results.Besides the processing of historical photograph, photogrammetry is combined with Artificial Intelligence to improve ways to search for architectural heritage in video material and to reduce the effort of manually examining them by the operator in the archive in terms of efficiency and time
Increasing the Segmentation Accuracy of Aerial Images with Dilated Spatial Pyramid Pooling
No full textThis thesis addresses the environmental uncertainty in satellite images as a computer vision task using semantic image segmentation. We focus in the reduction of the error caused by the use of a single-environment models in wireless communications. We propose to use computer vision and image analysis to segment a geographical terrain in order to employ a specific propagation model in each segment of the link. Our computer vision architecture achieved a segmentation accuracy of 89.41%, 86.47%, and 87.37% in the urban, suburban, and rural classes, respectively. Results indicate that estimating propagation loss with our multi-environment model reduced the root mean square deviation (RMSD) with respect to two publicly available tracing datasets
Deep Learning Based Automated Sports Video Summarization Using Yolo
No full textThis paper proposes a computationally inexpensive method for automatic key-event extraction and subsequent summarization of sports videos using scoreboard detection. A database consisting of 1300 images was used to train a supervised-learning based object detection algorithm, YOLO (You Only Look Once). Then, for each frame of the video, once the scoreboard was detected using YOLO, the scoreboard was cropped out of the image. After this, image processing techniques were applied on the cropped scoreboard to reduce noise and false positives. Finally, the processed image was passed through an OCR (Optical Character Recognizer) to get the score. A rule-based algorithm was run on the output of the OCR to generate the timestamps of key-events based on the game. The proposed method is best suited for people who want to analyse the games and want precise timestamps of the occurrence of important events. The performance of the proposed design was tested on videos of Bundesliga, English Premier League, ICC WC 2019, IPL 2019, and Pro Kabaddi League. An average F1 Score of 0.979 was achieved during the simulations. The algorithm is trained on five different classes of three separate games (Soccer, Cricket, Kabaddi). The design is implemented using python 3.7