Real Time Scene Segmentation of Video Data Using Histogramming Analisys
Autore
Fabio Sala - Politecnico di Milano - [1995-96]
Documenti
Abstract
This project addresses the problem of Real Time Scene Segmentation of video data. It involves the study of a method to automatically segment a video, recognise scene changes and collect a sequence of representative images that can ‘tell’ the story of the whole video footage.
The problem of scene segmentation has been developed using histogramming analysis. Various distance metrics has been investigated in order to obtain an intensity shift tolerant measure of scene change. This analysis has resulted in the development of a distance metric based on the vectorial distance between a geometric representation of consecutive histogram. The scene changes are detected using a window averaging threshold on the distance values in the sequence.
The feasibility of the adopted method for real-time segmentation has been studied in terms of speed and accuracy of the analysis process.
The method has been implemented with a graphic interface in order to obtain a powerful and flexible tool for scene segmentation analysis.
The problem of scene segmentation has been developed using histogramming analysis. Various distance metrics has been investigated in order to obtain an intensity shift tolerant measure of scene change. This analysis has resulted in the development of a distance metric based on the vectorial distance between a geometric representation of consecutive histogram. The scene changes are detected using a window averaging threshold on the distance values in the sequence.
The feasibility of the adopted method for real-time segmentation has been studied in terms of speed and accuracy of the analysis process.
The method has been implemented with a graphic interface in order to obtain a powerful and flexible tool for scene segmentation analysis.
Questa tesi è correlata alle categorie