This paper discusses LLE based nonlinear dimensionality reduction of gait images and its application in gait analysis and recognition. Firstly Gaussian derivative based gait segmentation method is given and it is better than classical background subtraction based method. Secondly self-similarity plot based gait alignment method is proposed. Then 1d LLE representation is used for gait analysis. Zero crossing, local maximum and minimum show gait cycle. The shape of 1d LLE representation shows dynamic feature of gait. 1d LLE representation is view independent. Finally the FFT module of 1d LLE representation is used for view independent gait recognition. Experiments results and its theory analysis show, LLE representation has the potential ability for view independent gait analysis and recognition.