Bag of Genres for Video Retrieval

Abstract

This paper presents a higher level representation for videos aiming at video genre retrieval. In video genre retrieval, there is a challenge that videos may comprise multiple categories, for instance, news videos may be composed of sports, documentary, and action. Therefore, it is interesting to encode the distribution of such genres in a compact and effective manner. We propose to create a visual dictionary using a genre classifier. Each visual word in the proposed model corresponds to a region in the classification space determined by the classifier's model learned on the training frames. Therefore, the video feature vector contains a summary of the activations of each genre in its contents. We evaluate the bag-of-genres model for video genre retrieval, using the dataset of MediaEval Tagging Task of 2012. Results show that the proposed model increases the quality of the representation being more compact than existing features.

Animation content in frame analysis

“…8, where the regions are classified into two groups: the regions with animation (Anim) and the regions without animation (Real). In the Anim group there are eight regions labeled by [1,2,5,6,9,10,13,14], where the rest of them are in the Real group ([3, 4, 7, 8, 11, 12, 15, 16]). Fig.…”

“…The visual information should be analyzed from various standpoints. Both static and dynamic visual information can be exploited using: time dependence [5]- [6], objects [7], motion [8], color, texture and similar descriptors [9]. The compressed domain may also be useful for such a task, as in [10] where the MPEG coefficients were applied.…”

An optimized unsupervised manifold learning algorithm for manycore architectures

“…Therefore, for the given example, the authority score A s ( q , 4) is equal to 14/4 2 . The time complexity for the computation of an authority score in Algorithm 1 (lines [11][12][13][14][15][16] is O ( k max 3 ), since the membership test in line 13 also needs to iterate through every object in N (q, k ) . On the other hand, the memoized version of the authority score computation given by Algorithm 3 is only O ( k max 2 ).…”

“…The first approaches were mostly based on low-level features, such as color, shape, and texture [24,39] . Motion patterns [1] and spatio-temporal properties [13] are some examples of other features used for encoding video content. Recently, mid-and high-level features based on local descriptors [3] and deep learning [46] have emerged as promising alternative solutions.…”