Pointwise Quantization Method for Improving Lifting Transform of G-PCC

In this paper, a pointwise quantization method that considers the reference characteristics of the G-PCC Lifting Transform is proposed. Depending on the level of detail of the Lifting Transform, the number of times each point is referenced in the prediction varies significantly. Reflecting these characteristics, we propose a method of allocating a proper quantization rate for each point according to influence of each point. Moreover, we also propose a rate-distortion optimized quantization for adaptive lambda values according to the influence of each point. The proposed method was implemented and evaluated on TMC13v14.0, the reference software of G-PCC. The proposed method achieves on average 0.7% YCbCr BD-rate gain under C1 condition.