Image Re-ranking Based On Dimension Reduction

In the age of information explosion, multimedia content is widely spread aroundthe world. As the rapid expansion demand of multimedia content has attracted moreand more scholars into the multimedia retrieval research. The content-based imagereranking has attracted prominent attention, it reranking the initial result which isfrom the text-based image ranking with the analysis of image’s visual content.However, the visual features are often in high-dimensional space which makes theranking very difficult and costly. Also, the traditional ranking methods are no longereffective for multimedia ranking. Dimensionality reduction is very helpful to tacklethese problems mentioned above.Multimedia ranking is the technology applied to ranking the content ofmultimedia, we analysis the state-of-art methods in the field and apply the mostsuitable one to the image reranking system.Feature dimensionality reduction is animportant step for data processing, which is used to reduce data’s dimensionalities inmany areas. In this paper, we apply dimensionality reduction to image searchreranking. As a supervised dimensionality reduction method, Linear DiscriminantAnalysis (LDA) performs well in classification applications, but is not the case forranking tasks. Firstly, it does not take the relevance degrees into consideration, whichis important for ranking problem. Secondly, owing to the supervised nature of LDA, aplenty of labeled samples are required, which are often costly and difficult to obtain.Therefore, based on LDA, we propose an improved method named Ranking LinearDiscriminant Analysis （RLDA） by using the relevance degrees as labels.Meanwhile, both labeled and unlabeled samples are utilized so that it is asemi-supervised approach.