Abstract: This paper proposes an improved graph transform-based 3D point cloud compression algorithm. The algorithm only generates one graph per processing block, so that each block will only produce a unique DC coefficient. In order to reduce the magnitude of the DC coefficient, we use the average of all the points in each block as the DC coefficient, and apply graph transform on the color value which has removed the average. In entropy coding, considering that a large amount of quantized AC coefficients is equal to zero, we use the Run-Level coding method to code non-zero coefficients. For DC coefficients, we use a predictive coding method which is similar to H.264 to coding the DC coefficients. The last coded AC coefficients and prediction residuals are coded by Huffman encoder. The experimental results show that the performance of proposed algorithm is better than the latest 3D point cloud compression algorithm.