In order to obtain the complex regions of cover image in adaptive steganography and improve the coherence between pixels, a novel adaptive steganography algorithm based on non-directional filter is proposed. The complex regions, which are difficult to be modeled and detected, are obtained by filter calculation on images through using a non-directional filter. The Gaussian low-pass filter is used for smoothing the complex regions, and enhance the coherence of different pixels in those regions and the cost function is established. Finally, according to the cost function, the messages are embedded by syndrome trellis codes. Experimental results demonstrate that the performance of anti-detection by the proposed algorithm is similar as that of the S-UNIWARD algorithm when the information embedding rate is small, and the proposed algorithm is better than the S-UNIWARD algorithm when the information embedding rate is large.