Hybridized Neural Network Based Approaches Used for Video Shot Boundary Detection

Abstract

Objective: The objective of this work is to detect video shot boundary which is an important phase in the domain of video processing. The applications such as indexing, retrieval and browsing of video are majorly depends on accurate detection of shot boundary so the detection is highly necessary in the area of video processing. Methods: Many researchers have contributed to the field of Shot Boundary Detection (SBD) to improve video processing throughout the previous decade. Some of the techniques have detected abrupt as well as gradual transitions with remarkable performance metrics values, but they compromised with the execution time and operating speed. Some of the techniques proposed and implemented the simpler approaches, but failed to detect the gradual transitions. Additionally, the effects like illuminations and camera motion lead to produce false results. Thus, taking these aspects into consideration, the novel approach for SBD has to be implemented. Findings: In this paper, fusion of transform and filtering at an early stage has been applied for reducing the false hits that occurs due to illumination noise. The fusion of transform with appropriate classifier enhances the detection performance. Another deep learning based technique with Rider bald eagle search optimization given the solution for the SBD. Novelty and applications:The proposed two approaches give the detection of abrupt and gradual transitions. It has been noticed that the said approaches found effective against the illuminations. To evaluate the performance it uses precision, recall and F1 score as performance metrics. The experiments are performed on latest TRECVID datasets and few open source video datasets.