This research is about the development of an automatic biometric identification system using gait energy images. Biometric recognition system has been extensively applied in computer vision applications such as airport security system, human surveillance system and criminal evidence gathering. However, some of these identification systems have their own drawbacks, therefore feedback from the subjects are needed in order to increase the sensitivity towards both environmental and physiological changes. Gait sequence consists of non-stationary data and can be modeled using a statistical learning technique. The proposed method for this research consists of 5 stages, namely pre-processing technique, feature extraction, dimensionality reduction, classifier and fusion. The first stage is pre-processing technique where the average silhouette images are calculated in order to capture the essential information in gait silhouette data. The second stage feature extraction process by applying Discrete Cosine Transform (DCT) and Local Binary Pattern (LBP). DCT defines a finite series of data points in terms of a sum of cosine functions fluctuating at different frequencies. On the other hand, LBP is also great texture descriptor because it compares each pixed with its neighbours by using certain threshold value. Both DCT and LBP were applied on the average silhouette to extract important gait features. The third stage is called dimension reduction process by using linear projection method. Principal Component Analysis (PCA) is applied to reduce the dimension of gait features. A linear projection method is employed in this stage to eliminate redundant features and remove noise from the gait image. In addition, this approach also increased the discriminating power in the feature space when dealing with low frequency information. Low dimensional feature distribution in the feature space is assumed as Gaussian, thus the Euclidean classifier can be considered as a better choice for the classification stage. In order to produce better gait recognition, Weighted Sum Fusion rule is applied to enhance class separation. This proposed algorithm is a model-free base which uses gait silhouette features for the compact gait image representation. This work has been evaluated with a benchmark CASIA dataset. From the experiment conducted, result shows that the best recognition rate is 97% when using 5 training images which are represented using 100 PCA coefficient, 50% DCT coefficient, LBP 3 x 3 neighbourhood, and Weighted Sum Fusion rule. The implementation of proposed method using Weighted Sum Rule fusion produced the best performance in terms of recognition.
