Applying Image Pre-processing Techniques for Appearance-Based Human Posture Recognition: An Experimental Analysis

This paper investigates that loose clothing such as wearing dresses and human body shapes create individual eigenspaces and, as a result, conventional appearance-based method cannot be effective for recognizing human body postures We introduce a dress effect due to loose clothing and a figure effect due to various human body shapes in this particular study This study particularly proposes an image pre-processing by ‘Laplacian of Gaussian (LoG) ' filter over input images and a ‘mean posture matrix ' for creating an eigenspace in order to overcome the preceding effects We have tested the proposed approach on various dress environments and body shapes, and robustness of the method has been demonstrated.