A novel subband coding approach is presented in this paper.It uses the 2-D all phase discrete cosine sequency filter(APDCSF) to decompose original image into subbands;the low-pass subband image is directly multi-subasampled with a two-dimensional cross structure and interpolated with the two-dimessional rotating multi-interpolation filting based on the all phase discrete inverse cosine sequency filter(APDICSF).As to the high-pass subband images,the image primitives such as the edge and line feature are extracted by histogram automatic thresholding.Different compression coding techniques are performed respectively for the features in different subbands,and the original image can be reconstructed by convex projection.This approach with using the APDCSF can eliminate the block effects in traditional discrete cosine transform(DCT) transform coding.Compared with the subband feature coding based on wavelet transform,its computational complexity is much less and its compression rate is higher,furthermore the subjective visual effect is better than DCT transform coding.It is suitable especially for low bit rate image compression.