Variable bandwidth mean shift for video image segmentation

In order to extract motion objects effectively from the video sequence, a new nonparametric estimation method called variable bandwidth mean shift (VBMS) algorithm is applied to realize image smoothing and segmentation, which removes the small texture and preserves the structural features of the image objects. The intersection of confidence intervals (ICI) rule and the sample point density estimator are employed in VBMS algorithm to determine the smoothing bandwidth of each pixel in the video image. With the probability density function (PDF) estimator and the bandwidth parameter, the mean shift vector is obtained. Along the direction of this vector, image is smoothed and the pixels are clustered to several blocks. At the same time, motion blocks are detected and the features of the affine motion parameters are analyzed along the time axis. Image blocks with similar motion parameters assemble to an integrated part and the motion objects are extracted from the video images. The experiments verified the effectiveness of the proposed method.