With this rapid increase in generating and collecting images, image mining and knowledge discovery attract so much attention. In this paper, we present a framework for extracting knowledge from sequence of images in the form of natural languages. In another view point, this process summarizes the enormous visual information of sequence of images into some simple and small sentences with the aim of automatic traffic regulations checking. Sequences of images have been manually extracted from a video of moving vehicles in cities. The structure of the system composed of Image Analysis and Knowledge Processing modules. Image Analysis module works on raw images and tries to extract a set of records consisting of spatial-temporal characteristics of moving objects in the scene. These records are passed to knowledge processing phase for further processing, information transformation and summarization. Knowledge processing phase consists of two modules: Trajectory Detection and Fuzzy Rule base. A new similarity measure based on Longest Common Subsequence (LCS) is proposed and used in trajectory detection module. This module tries to form traffic behavioral classes. In fuzzy Rule Base module, traffic regulations have been modeled using fuzzy rules and fuzzy inference used for checking normality/abnormality of the actions in the scene. The output of the system in the form of Natural Languages clearly shows what happened in the scene.