Multimedia Retrieval – Perspectives and Challenges

Abstract

In this chapter, a new feature descriptor, local mesh correlation histograms (LMeCH) is proposed for content-based image retrieval (CBIR). The LMeCH integrates the local mesh patterns (LMeP) and grayscale joint histogram. Firstly, the LMeP features are extracted from the image and then the joint histogram is constructed between the LMeP and grayscale value of center pixel. The process of joint histogram is able to extract the efficient image features from the databases. The retrieval performance of the proposed method is tested on two bench mark OASIS-MRI and NEMA-CT biomedical image databases. The experimental results show a significant improvement in terms of precision, recall, average retrieval precision (ARP) and average retrieval rate (ARR) when compared with other standard image retrieval approaches on the same database.

Abstract

This chapter presents a text mining tool for scientific publications that allows the extraction of textual segments (section, paragraph, sentences, etc.) from a large corpora according to a set of semantic categories (results, methods, hypothesis, etc.). The extracted information is grouped according to their semantic affiliation which allows to obtain an organized textual representation called multi-document synthesized sheets. The automatic construction of these synthesized sheets is realized by semantically annotating documents according to a set of semantic categories. In fact, the annotation task is performed automatically using the Contextual Exploration processing (EC). It is a computational linguistic method based on a set of linguistic markers associated with semantic categories.

Abstract

Advanced information retrieval systems face a great challenge arising from the emergence of massive and multi-modal data, including images, texts, video, audio and etc.
One of the most important problems in this field is to accomplish effective and efficient search across various modalities of information. Given a query from one modality, it is desirable to retrieve semantically relevant answers in all the available modalities. This chapter first briefly reviews related works, including uni-modal information retrieval, multi-modal information retrieval, and cross-modal information retrieval. For crossmodal retrieval models, this chapter gives an introduction to manifold alignment-based model (MAM), naive topic correlation model (NTC), and semantic topic correlation model (TCM) and their correspondence mapping techniques, particularly semantic topic correlation mapping. An extension of TCM applied to retrieve information in Chinese language is also introduced in this chapter.

Abstract

This chapter presents a performance comparison of the Error Diffusion Block Truncation Coding (EDBTC) feature for color image retrieval and classification. In these approaches, the image retrieval and classification employ the feature descriptor derived from the EDBTC compressed data stream. Firstly, a color image is decomposed using EDBTC scheme to produce two new image representations, namely color quantizer and bitmap image. Two image feature descriptors, called Color Histogram Feature (CHF) and Bit Pattern Histogram Feature (BHF), can be subsequently generated from the EDBTC color quantizer and its corresponding bitmap image, respectively, without performing the decoding process. The similarity degree between two images is simply measured with the similarity distance score of their feature descriptor. In this chapter, the effectiveness of EDBTC feature descriptor is quantitatively examined and compared in the RGB color space as well as in YCbCr color channel. As reported in experimental result, the proposed method outperforms the former existing schemes in the image retrieval and classification tasks. It has shown that the EDBTC performs well in image compression domain, in addition, it also offers an effective and efficient way for performing image retrieval and classification.