This study investigates the impact of Histogram of Oriented Gradients (HoG) on gait recognition. HoG is applied to four basic gait representations,
i.e.
Gait Energy Image (GEI), Gait Entropy Image (GEnI), Gait Gaussian Image (GGI), and newly developed Gait Gaussian Entropy Image (GGEnI). Hence their corresponding secondary gait representations, Gradient Histogram Gait Images (GHGI), are generated. Due to the nature of HoGs, the secondary gait representations contain rich information of images from different scales and orientations. The optimized HoG parameters are investigated to establish appropriate parameter settings in the HoG operations. Evaluations are conducted by using Support Vector Machines (SVM) as classifier swith CASIA dataset B. Experimental results have shown that HoG associated secondary representations are superior to the originally basic representations in gait recognition, especially in case of coping with appearance changes, such as walking with bag and walking with coat when using normal walking samples in training. GHGIs have increased the gait recognition rate approximately of 17% to GEI, 12% to GEnI, 24% to GGI and 20% to GGEnI.