Toward Robust Action Retrieval in Video

Visual recognition and interpretation of human-induced actions and events are among the most active research areas in computer vision, pattern recognition, and image understanding [4]. In this work, we develop such a framework for robustly recognizing human actions in video sequences. The contribution of the paper is twofold. First a reliable neural model, the Multi-level Sigmoidal Neural Network (MSNN) as a classifier for the task of action recognition is presented. Second we unfold how the temporal shape variations can be accurately captured based on both temporal self-similarities and fuzzy log-polar histograms. Neural network classifier has many advantages over other competitive machine learning classifiers. Some of these advantages include the high rapidity, easiness of training, realistic generalization capability, high selectivity and great capability to create arbitrary partitions of feature space. However, the neural model, in the standard form, might have low classification accuracy and poor generalization properties because its neurons employ a standard bi-level function that gives only two values (i.e., binary responses) [2]. To relax this restriction and allow the neurons to generate multiple responses, a new functional extension for the standard sigmoidal functions should be developed. This extension is termed the Multi-level Activation Function, and the model that uses this extension is termed the Multi-level Sigmoidal Neural Network (MSNN). It is straightforward to derive a multi-level version from a given bi-level standard sigmoidal activation function f (x) as follows