Nonlinear Dynamic Texture Analysis and Synthesis Model

Dynamic textures are categorized into linear dynamic texture and Non-linear dynamic texture. Linear dynamic texture exhibits high temporal self-similarity. A lot of work has been performed to synthesize dynamic textures. Reproducing the visual realism of the real world is a major goal for computer graphics and image processing, and most of the real world dynamic textures are nonlinear in nature. Examples of nonlinear real world motions are playing child, moving flag etc. In this paper, the innovative method for analyzing and synthesizing linear and nonlinear dynamic textures is proposed. To capture nonlinear motion, dynamic textures are modeled by dynamic texture units. The dynamic texture units’ parameters are learnt at the same time. Daubechies discrete wavelet transform, and different color coding i.e. YCbCr, YUV, and YIQ are used to get better visual quality and compact representation. It exploits the spatial, temporal, and chromatic correlations amongst the pixels to get the more compact model parameters. The proposed model has the predictive power and can be used to generate infinite synthetic sequence with negligible computational cost. To reduce time complexity, modified LDS (Linear Dynamic System) algorithm is implemented on GPU. Testing is done on various dynamic textures. It reconstructs the infinite dynamic texture frames with promising visual quality, fewer coefficients, and less time complexity. The testing result shows the proposed dynamic system along with Daubechies discrete wavelet transform, and YIQ color coding achieves a higher compact model with an acceptable visual quality than the available LDS and Fourier descriptor model. A compact representation for dynamic texture synthesis to generate more number of frames than the available, is very useful for embedded devices, which have a limited memory and computational power, such as tablets, mobile phones, etc.