According to deterministic character and geometric property of chaotic system, the predictive method of neighborhood selection and the approach of neighborhood selection based on Delaunay triangulation were presented, and a new orthogonal local projective algorithm was derived by improvement of local projective method for noise reduction in chaotic time series, and rules of selection of optimal parameters were suggested. The method was successfully applied to extract weak signals in Lorenz chaos， that provided signal to noise ratio not lower than -80 dB. Computer simulation shows that this projective filtering method has great stability and reliability, and is effective for extracting weak signals in chaos.