Lei Zhang








Video Segmentation using a Spatio-temporal Conditional Random Field

One of my early projects is on video segmentation  using Conditional Random Fields (CRFs). CRF has been showed to outperform Markov Random Fields (MRFs) in several image segmentation and natural language processing tasks. However CRFs have not been widely applied for video segmentation at that time (2006). In this work I developed a spatio-temporal CRF (STCRF) that encodes both the spatial and temporal relationships within a local neighborhood. In addition, we kept the discriminative nature of CRF modeling. In contrast, some other CRF models for video sequence segmentation actually model the likelihood of image measurements, very similar as the traditional MRF modeling that is actually a generative model. 


Example results:

We have tested experiments on several public available video sequences. Here are some examples of our segmentation results. 

Description: D:\Lei Zhang\MyWebpage\index_files\STCRF_result.PNG

Description: D:\Lei Zhang\MyWebpage\index_files\STCRF_result1.PNG

Figure 1. Examples of video segmentations using the STCRF model

Related publications:

§   Lei Zhang and Qiang Ji. "Segmentation of Video Sequences using Spatial-temporal Conditional Random Fields", in IEEE Workshop on Motion and Video Computing (WMVC 2008)