A5102978003- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Advanced Neural Network Applications
- AI in cancer detection
- Video Surveillance and Tracking Methods
- Radiomics and Machine Learning in Medical Imaging
- Multimodal Machine Learning Applications
- Image and Video Quality Assessment
- Advanced Image Fusion Techniques
- Medical Image Segmentation Techniques
- Face and Expression Recognition
- Digital Radiography and Breast Imaging
- Parasites and Host Interactions
- Data Management and Algorithms
- Voice and Speech Disorders
- Computational Geometry and Mesh Generation
- Digital Imaging for Blood Diseases
- Speech and Audio Processing
- Advanced Text Analysis Techniques
- Neural Networks and Applications
- Sentiment Analysis and Opinion Mining
- Optimization and Packing Problems
- Sparse and Compressive Sensing Techniques
- Blind Source Separation Techniques
- Parasite Biology and Host Interactions
Huawei Technologies (United Kingdom)
2020-2022
Universidade do Porto
2018-2022
INESC TEC
2017-2022
Huawei Technologies (Sweden)
2021
Institute for Systems Engineering and Computers
2019
Instituto de Engenharia de Sistemas e Computadores Investigação e Desenvolvimento
2018
University of California, San Diego
2010-2016
Telefonica Research and Development
2014
Universidade de São Paulo
2008
Centro Universitário do Norte Paulista
2006
The problem of joint modeling the text and image components multimedia documents is studied. component represented as a sample from hidden topic model, learned with latent Dirichlet allocation, images are bags visual (SIFT) features. Two hypotheses investigated: that 1) there benefit to explicitly correlations between two components, 2) this more effective in feature spaces higher levels abstraction. Correlations canonical correlation analysis. Abstraction achieved by representing at...
The problem of cross-modal retrieval from multimedia repositories is considered. This addresses the design systems that support queries across content modalities, for example, using an image to search texts. A mathematical formulation proposed, equating isomorphic feature spaces different modalities. Two hypotheses are then investigated regarding fundamental attributes these spaces. first low-level correlations should be accounted for. second space enable semantic abstraction. Three new...
<h3>Importance</h3> Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography accuracy by reducing missed cancers and false positives. <h3>Objective</h3> To evaluate whether AI can overcome interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. <h3>Design, Setting, Participants</h3> In this diagnostic study conducted between September 2016 November 2017, an...
Two limitations hamper performance of deep architectures for classification and/or detection in medical imaging: (i) the small amount available data, and (ii) class imbalance scenario. While millions labeled images are today to build tools natural scenes, annotated data automatic breast cancer screening is limited a few thousand images, at best. We address these with method augmentation, based on introduction random elastic deformations mammograms. validate this three publicly datasets. Our...
This paper reports on the NTIRE 2021 challenge perceptual image quality assessment (IQA), held in conjunction with New Trends Image Restoration and Enhancement workshop (NTIRE) at CVPR 2021. As a new type of processing technology, algorithms based Generative Adversarial Networks (GAN) have produced images more realistic textures. These output completely different characteristics from traditional distortions, thus pose for IQA methods to evaluate their visual quality. In comparison previous...
This paper addresses the problem of document retrieval based on sentiment polarity criteria. A query natural spontaneous speech, expressing an opinion about a certain topic, is used to search repository documents containing favorable or unfavorable opinions. The goal retrieve whose opinions more closely resemble one in query. semantic system speech transcripts augmented with information from full-length text articles. Posterior probabilities extracted articles are regularize their...
Nowadays, fractal analysis has been successfully applied to digital speech processing, particularly for word and phoneme segmentation, which represents one of the fundamental steps in automatic recognition systems. The practical use this purpose should match two principles: low computational cost, allow real-time, accuracy results, order produce a satisfactory sending correct data classifier. Aiming at meeting these requirements, work proposes technique segmentation based on dimension, is...
Este artigo apresenta os resultados do inquérito epidemiológico que estudou a ocorrência e distribuição da filariose bancroftiana no Distrito de Cavaleiro, Município Jaboatão dos Guararapes, Pernambuco. O desenho estudo empregado foi o seccional. Foram analisados 9.520 indivíduos população residente nos 12 bairros compõem distrito. Deste total, detectou-se 213 microfilarêmicos (2,2%). Cerca 91,7% pesquisados apresentaram casos infecção filarial, com prevalências variando 0% 5,15%. A...
We present an algorithm to distinguish between pathological and normal human voice signals based on discrete wavelet transforms (DWT) support vector machines (SVM). The former is used for time-frequency analysis provides quantitative evaluation of signal characteristics. latter the final classification. technique leads adequate larynx pathology classifier with over 95% classification accuracy.
Convolutional Neural Networks (CNN) have become the gold standard in many visual recognition tasks including medical applications. Due to their high variance, however, these models are prone over-fit data they trained on. To mitigate this problem, one of most common strategies, is perform augmentation. Rotation, scaling and translation operations. In work we propose an alternative method rotation-based augmentation where rotation transformation performed inside CNN architecture. each...
A key to the generalization ability of Convolutional Neural Networks (CNNs) is idea that patterns appear in one region image have a high probability appearing other regions. This notion also true for spatial relationships, such as orientation. Motivated by fact early layers CNNs distinct filters often encode same feature at different angles, we propose incorporate rotation equivariant prior these models. In this work, regularization strategies capture approximate equivariance were designed...
We present an engineered version of the divide-and-conquer algorithm for finding closest pair points, within a given set points in XY-plane. For this we show that only two pairwise comparisons are required combine step, each point lies 2 delta-wide vertical slab. The correctness is shown all Minkowski distances with p>=1. also empirically that, although time complexity still O(n lg n), reduction total number leads to significant execution time, inputs size sufficiently large.
Image hash codes are produced by binarizing the embeddings of convolutional neural networks (CNN) trained for either classification or retrieval. While proxy achieve good performance on both tasks, they non-trivial to binarize, due a rotational ambiguity that encourages non-binary embeddings. The use fixed set proxies (weights CNN layer) is proposed eliminate this ambiguity, and procedure design sets nearly optimal hashing introduced. resulting hash-consistent large margin (HCLM) shown...
Image quality assessment (IQA) forms a natural and often straightforward undertaking for humans, yet effective automation of the task remains highly challenging. Recent metrics from deep learning community commonly compare image pairs during training to improve upon traditional such as PSNR or SSIM. However, current comparisons ignore fact that content affects only occur between images similar content. This restricts diversity number model is exposed training. In this paper, we strive enrich...