At the 14th JVET meeting, several contributions proposed to simplify the chroma scaling factor derivation to reduce the dependency between luma component and chroma component, and a CE was set up to further study it. This contribution proposes to use CTU neighboring reconstructed samples to derive chroma scaling factor to solve the dependency issue. It is reported that the performance impact for {Y, U, V} is {0.00%, -0.07%, -0.06%} for RA and {0.02%, 0.00%, -0.20%} for LDB in the case of using 4 left and 4 up neighbors, and {0.00%, -0.09%, -0.13%} for RA and {-0.01%, -0.07%, -0.19%} for LDB in the case of using 16 left and 16 up neighbors if the CTU-level derivation method is applied only in case of single tree. And by additionally applying the proposed method in the case of dual tree, and enabling the chroma scaling in dual tree, the total coding performance is {0.05%, -0.49%, -0.65%} for AI, {0.03%, -0.65%, -0.66%} for RA and {-0.02%, -0.59%, -0.51%} for LDB in the case of using 4 left and 4 up neighbors, and {0.04%, -0.43%, -0.59%} for AI, {0.03%, -0.67%, -0.67%} for RA and {0.02%, -0.47%, -0.53%} for LDB in the case of using 16 left and 16 up neighbors.