In order to improve the coherent among different pixels, a spatial image adaptive steganography algorithm is proposed based on the framework of minimizing the embedding distortion function. The Daubechies wavelet construction filter is used to predict the residual weight of cover and the cost value is obtained. A mean filter is used to smooth the cost value. The messages combined with syndrome trellis coding are embedded. Two different image features are used for anti-steganography analysis. Experimental results demonstrate that the anti-detection ability of the proposed algorithm is similar to the Hill-pass, Low-pass, Low-pass (HILL) steganography algorithm when the payload is small and is significantly better than the mainstream steganography algorithms such as wavelet obtained weights and spatial-universal wavelet relative distortion (S-UNIWARD ) when the payload is large.