A homomorphic system based approach for extracting invariant features of object images is proposed in this paper. By this approach, the statistical values such as mean, variance, skewness, and kurtosis are calculated from the Walsh-domain intensity histogram of the object image. It is found that the statistical characteristics of this histogram are invariant under different environments of the object image such as translation, rotation, scale change, illumination change, and noise corruption. Hence these values can be used as features of the object image and applied to check whether the object is locked on or not during the tracking phase in an image tracking system. By experimental results, we find that these features keep close or better performance compared with the well known invariant moments methods for TV or FILR images, respeectively.
