Article Preview
Top1. Introduction
Context and Problems: In audio-visual document retrieval, providing a better description is crucial to represent audio-visual document segment with a set of concepts reflecting the content. In this context, several efforts were made in order to extract important descriptions from the content. Although information retrieval techniques were adapted to explore the audiovisual documents, the problem of user dissatisfaction persists. To achieve this purpose, the extraction of descriptions requires analysing the content in order to present knowledge and thus perform a better interrogation. The existing difficulties are due to the inappropriate used data sources. In fact, a bad choice of sources may be the cause of irrelevance of the information returned during a research process. Indeed, the ambiguity of the link between the different descriptions can complicate the interrogation process. Thus, a fairly complete and consistent description of audio-visual documents is necessary in this procedure because of the dearth of the automatic description processes for such resources.
Objective: In this paper, we introduce an approach for the description of the cinematic audio-visual documents through a semiotic description as they provide a rich knowledge representation to the content. In fact, the semiotic descriptions were defined in the literature as those that enable exploring the meaning and interpreting the various kinds of knowledge in the content (Martin, 2005). Peter Stokinger, in (P Stockinger, 2011), (Peter Stockinger, 2013) and (P Stockinger, 2003), defined audiovisual semiotics as a theoretical framework that follows a methodology in the description and interpretation of film content to analyse and make explicit annotations and even ideas that bring meaning to audio-visual plans. In this regard, semiotics description is considered, in this work, as the process of providing the significance of the audiovisual document in order to link the semantics and the various meanings conveyed in the content of the audio-visual document. Our research contributions are based on the studies carried out by Peter Stockinger. This work shows the importance of describing and structuring the content of filmic document using semiotic descriptions. The introduced approach is based on three types of documentations related to the cinematic document: the superposed text on the image, the script used for the writing phase and the synopsis of the movie.
The main objective of this research work is to detect the semiotic and thematic description related to each filmic segment. In fact, semiotic description represents the relevant description reflecting the content such as, dominant theme, themes of discourses, taxeme and specified themes. In this context, we propose an approach to segment, extract and identify these descriptions. The major contributions of this paper are:
- -
Temporal and semantic segmentation of the movie document using the textual segmentations of the script of the movie and the superposed text on the image.
- -
Extraction of the generic description by employing the standard Dublin core.
- -
Extraction of semiotic description by adapting the latent Dirichlet allocation (LDA) model.
- -
Identification of descriptions related to each segment of the film.
Organization of the Paper: The rest of the paper is organized as follows. Section 2 presents an overview of the related works presented in the literature dealing with the extraction of descriptions in audio-visual documents. Section 3 describes the used background of the semiotic description. Section 4 introduces the developed approach for the extraction and the identification of external and semiotic description. Section 5 depicts the datasets utilized to evaluate the performance of the suggested approach and to carry out the presentation of the various performed experiments. Section 6 discusses the limitation of the proposed approach and some futures scope. The final section presents our conclusions and recommendations for future research.