Combinando imagem e som para detecção de transições em vídeos digitais

• Ano de defesa: 2008
• Autor(a) principal: Fonseca, Marcilene Scantamburlo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Programa de Pós-Graduação em Computação; Computação
• Palavras-chave em Português: Ciência da computação; Computação visual; Sistemas nervoso autonômos; Vídeos; Processamento de imagem; Análise de imagens; Recuperação de imagens; Segmentação de vídeo; Teoria Bayesiana; Recuperação de áudio; Estatística; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO::COMPUTABILIDADE E MODELOS DE COMPUTACAO
• Link de acesso: https://app.uff.br/riuff/handle/1/17887

The large and growing volume of digital applications, such as video indexing, retrieving, analysis, compression, storing and data transmission, have being accomplished based on scene transition detection, sometimes called video segmentation. Since it is a basic step and it is far from a complete solutions, it is also a open area, needing much more for investigations. Although scene classification by human concerns several attributes, the use of color, shape and texture are the more frequently used on computational models. On the other hand, researches on digital sound processing and voice signal have gotten much attention, mainly in multimedia systems. It is well known, however, that each attribution on processing of scene analysis can relies on several features at the same time, as visual as sonorous. This dissertation proposes has as main goal the analysis of the results in automatic detection of scene transition using traditional visual features combined with sounding features to the scene. This detection regards the color aspects (based on color histogram), shape aspects (based on edge map histogram) and sound aspects, which is accomplished with 7 sound characteristics, namely: short time average energy, zero-crossing rate, energy band ratio, delta spectral magnitude, root mean square of square sum of signals, high sounds and low value characteristics ratios. A Bayesian network model is used on this analysis and the results shows that the inclusion of sound in the model combining color and shape can improve the detection performance.