Desenvolvimento de um sistema para detecção de objetos de uma imagem de referência estática utilizando método estatístico paramétrico para aplicações de Chroma Key

Abstract

The traditional Chroma Key method employs a technique called Key Mean Clustering, which demands a substantial amount of storage through buffers, as well as system speed and complexity for hardware implementation. This research project presents the MATLAB validation of the development of a low-complexity implementation system for Chroma Key applications. This system performs the extraction of objects from a static reference image with a green background using image segmentation based on a parametric statistical model. This model utilizes an average image of color components as a reference to classify each image element. For final validation, 120 images were used for testing, consisting of 37 photos with a green background and 83 photos with the object to be detected present. Additionally, performance evaluation was conducted using the following metrics: precision, recall, Fmeasure, and confidence factor. Utilizing the RGB color space, a maximum accuracy of 100% and a minimum of 99.4% were achieved. In the case of the YCbCr color space, the maximum accuracy attained was 100%, and the minimum was 99.9%. Through statistical inference testing, it can be concluded that there are no statistically significant differences between the two color spaces.