A plurality of cameras is placed so that those optical axes become parallel to the Z-axis and are in relative position t on the XY plane. Multi-view video compression coding apparatus determines depth distance Z and unit normal vector (nx, ny, nz) of the arbitrary point in the surface of the object, moves the first block by disparity vector based on t and Z, transforming the first block by linear transformation matrix based on t, Z and n, then, matching the first block to block of the second picture. Then, while changing Z and n arbitrarily, the second block of the second picture which is most similar to the first block is searched for. And the apparatus derives a prediction error between the first block and the second block, codes the prediction error, and adds Z and n to coded data.