Most image compressed sensing algorithms improve the reconstruction quality by utilizing the correlation of parent-child wavelet coefficients. However, few people study the compressed sensing based on the fraternal relationship of the high-frequency coefficients. In this paper, a Bayesian-based image compressed sensing algorithm using joint reconstruction of high-frequency wavelet coefficients is proposed. Firstly, the high-frequency coefficients of the horizontal, vertical and diagonal directions in the same scale are sampled separately when executing compressed sensing. Then, a hierarchical Bayesian model is presented and the correlation is used when reconstruction is performed. Experimental results show that our proposed algorithm has higher image reconstruction quality than the existed MCS.