Modelos de Aprendizaje Profundo para el Procesamiento y Clasificación de Imágenes y Vídeo

Resumo

Motivated by the success of DL-based models in image and video problems, in this dissertation, we develop DL-models for challenging image and video formation and interpretation tasks. These are image and video SR, BID, threat detection in PMMWIs and mitosis detection in Whole-Slide Images (WSIs). In this thesis, one common point to all contributions is the use of domain knowledge to improve the solution by developing and applying specialized architectures, regularizations and restrictions. In the next subsections, we present the tasks that we have addressed in this dissertation. Next, we provide a brief introduction to the main DL-based models used in this dissertation: CNNs and Generative Adversarial Networks (GANs). Finally, we present the objectives of this work and the structure of the remainder of this dissertation.