We propose a novel texture feature extraction technique based on coefficients' co-occurrence histogram of discrete wavelet frame transformed image, which capture the information about relationship between each high frequency subband and the low frequency subband of the decomposed image at the corresponding level. It is not independently utilizing the information of each subband coefficient. The classification performance is analyzed using the k-NN classifier. And the experimental results demonstrate the effectiveness of our proposed texture feature in achieving the improved classification performance. Comparisons with the Gabor filter and a recently proposed approach are also provided.