Multimedia data accessibility depends on a precise indexing, involving a computational cost. This paper proposes a new fast method of segmentation and indexing in order to fill out in an automatic way several MPEG7 fields (e.g. camera and objects movement). In order
to accelerate segmentation process, we exploit most of the information contained in MPEG1-2 flow; the decompression is restricted to entropic decoding and inverse quantization, the estimation of the camera movement is obtained from MPEG1-2 motion prediction. Segmentation in homogeneous color zones is obtained by a "split and merge" algorithm improved by a B-Splines active contour segmentation regularization.