Abordagem para a estilização de vídeos digitais.

Abstract

Non-photorealistic rendering is a research area that lies in the intersection of Computer Graphics, Digital Image Processing and Art. Using images and videos captured from real world scenes stylized and abstracted representations of reality are generated to visually communicate ideas or concepts to a human observer. Within this context, the present dissertation propose and validate an automatic approach to photorealistic video cartoonization. The approach is composed by the following steps: (1) Edge preserving smooth filtering (Adaptative Bilateral Filter) to smooth video frame regions, (2) Color quantization for region segmentation and to reduce the number of colors in the video, (3) Determination of connected components to reduce flickering effects and temporal coherence enhancement of the rendered video, (4) Edge enhancement using an edge detector, and (5) Saturation and contrast adjustments in the video frames in order to improve visualization. As a result a prototype implementing the proposed approach has been produced. The prototype parameters were experimentally tested and validated using quantitative and qualitative tests consisting of an assessment of the opinion of volunteers on videos extracted from a public digital database. Statistical analyses were conducted upon the collected data in order to indentify parameter combinations and techniques that yielded the best results for the participants of the experiment, thus indicating the best operating conditions for the successful use of the proposed approach.