Searching in video collections using sketches and sample images – the cineast system

Abstract

With the increasing omnipresence of video recording devices and the resulting abundance of digital video, finding a particular video sequence in ever-growing collections is more and more becoming a major challenge. Existing approaches to retrieve videos based on their content usually require prior knowledge about the origin and context of a particular video to work properly. Therefore, most state of the art video platforms still rely on text-based retrieval techniques to find desired sequences. In this paper, we present Cineast, a content-based video retrieval engine which retrieves video sequences based on their visual content. It supports Query-by-Example as well as Query-by-Sketch by using a multitude of low-level visual features in parallel. Cineast uses a collection of 200 videos from various genres with a combined length of nearly 20 h.