A Framework for Multimedia Event Classification With Convoluted Texture Feature

Background: The proposed work uses two approaches as its background. They are (i) LBP approach (ii) Kirsch compass mask. : Texture classification plays a vital role in object discrimination from the original image. LBP is majorly used for classifying texture. Many filtering based methods co-occurrence matrix method, etc., were used, but due to the computational efficiency and invariance to monotonic grey level changes, LBP is adopted majorly. Second, as Edge plays a vital role in discriminating the object visually, Kirsch compass mask was applied to obtain maximum edge strength in 8 compass directions which has the advantage of changing the mask according to users own requirement than any other compass mask. Objective: The objective of our work was to extract better features and model a classifier for the Multimedia Event Detection task. Methods: The proposed work consists of two steps for feature extraction. Initially, an LBP based approach is used for object discrimination, later, convolution is used for object magnitude determination using Kirsch Compass mask. Eigenvalue decomposition is adopted for feature representation. Finally, a classifier is modelled using a chi-square kernel for the event classification task. Results: The proposed event detection work is experimented using Columbia Consumer Video (CCV) dataset. It contains 20 event based videos. The proposed work is evaluated with other existing works using mean Average Precision (mAP). Several experiments have been carried out to evaluate our work, they are LBP vs. non-LBP approach, Kirsch vs. Robinson compass mask, Kirsch masks angle wise analysis, comparison of above approaches are performed in a modeled classifier. Two approaches are used to compare the proposed work with other existing works.They are (i) Non Clustered Events (events were considered individually and one versus one strategy was followed) (ii) Clustered Events (some events were clustered and followed one vs. all strategy and remaining events were non-clustered). Conclusion: In the proposed work, a method for event detection is described. Feature extraction is performed using LBP based approach and Kirsch compass mask for convolution. For event detection, a classifier model is generated using the chi-square kernel. The accuracy of event classification is further increased using clustered events approach. The proposed work is compared with various state- of- the- art methods and proved that the proposed work obtained outstanding performance.

How Unlabeled Web Videos Help Complex Event Detection?

The lack of labeled exemplars is an important factor that makes the task of multimedia event detection (MED) complicated and challenging. Utilizing artificially picked and labeled external sources is an effective way to enhance the performance of MED. However, building these data usually requires professional human annotators, and the procedure is too time-consuming and costly to scale. In this paper, we propose a new robust dictionary learning framework for complex event detection, which is able to handle both labeled and easy-to-get unlabeled web videos by sharing the same dictionary. By employing the lq-norm based loss jointly with the structured sparsity based regularization, our model shows strong robustness against the substantial noisy and outlier videos from open source. We exploit an effective optimization algorithm to solve the proposed highly non-smooth and non-convex problem. Extensive experiment results over standard datasets of TRECVID MEDTest 2013 and TRECVID MEDTest 2014 demonstrate the effectiveness and superiority of the proposed framework on complex event detection.