The concept and method of decomposable Markov network (DMN) are sdudied, and the role of DMN in data mining from sequence images is analyzed. The construct of DMN is straight utilized as the evidence of inference or decision for problem solution and enlargement of application possibility in DMN. For example of the factual video images with traffic rule violation, we deeply investigate several graininess (nodes) which are used in constructing DMN between video images, assess its applicability for our problem, and detect abnormality in them by analyzing construct of DMN for finding a interesting mode. It is showed that the method is feasible and effective according to the result of simulation. The researches indicate that the DMN may apropos reveal abstract adjacent relations existing in data, has capabilities of showing knowledge and logic reasoning, and also is important method of pattern recognition.