In this talk, we address tensor completion problem from sparsely observations corrupted by impulse noise. Based on the MCP function, we propose a non-convex model, which minimizes a weighted combination of tubal nuclear norm and MCP term. Although this problem can be solved by block coordinate descent (BCD) method, a direct application of BCD method often leads to difficult nonconvex subproblems. To address this issue, we propose to convexify the subproblems through a linearization technique as done in the difference of convex functions algorithm (DCA). Then we establish a block proximal Difference of Convex Algorithm (BPDCA) to solve it. We prove that the approximation sequence converges globally to a stationary point of the proposed model. Numerical experiments verify the effectiveness of our model and algorithm.