Correlation denoising could distinguish noisy and signal coefficients by the correlation degree and thus retain more right edge. But there is a strict requirement to unerring locations. The nonsubsampled contourlet (NSCT) is a shift-invariant directional multiresolution image representation and overcomes the disadvantage of wavelet, the nonoptimal basis for one-dimensional singularity. For the shift-invariance, it could not only preserve more edge details than contourlet but also satisfy the above requirement of correlation denosing. Thus, to combine threshold denoising, we present a novel inter-scale correlation and threshold combination denosing model of NSCT (ISCTC-NSCT). The simulation results have shown that the performance of the above method is superior both in signal to noise ratio (SNR) and edge preservation.