To resolve the problems of high computational complexity, low anti-noise ability and the drifting of pixel position, a scale invariant feature algorithm based on causality is proposed in this paper. Firstly the Gauss smoothing image is built up by Gaussian convolution with the original image. Then, the Harris corners are extracted as candidate features both in the original and the Gauss image. Finally, the stable scale invariant features are acquired by projection from the original image to the Gauss image. The experimental results indicate that this algorithm is concise, fast, efficient with strong anti-noise ability, and provides a basis for subsequent visual processing.