Abstract
This paper addresses the challenge of predicting the compression performance of the computationally complex Better Portable Graphics (BPG) image codec without executing its demanding encoding process. We propose a novel methodology that leverages a linear mapping between the bits per pixel (bpp) of JPEG and BPG compressed images. The core of the approach involves determining optimal pairs of JPEG Quality Factor (QF) and BPG quantization parameter (Q) that maximize bpp correlation, and deriving corresponding linear mapping laws. These laws, established once on a training set, enable BPG bpp prediction for any new image using only a fast JPEG compression. The method was rigorously validated on a diverse dataset including color, grayscale, and infrared images from both the same and different sensor systems. Results demonstrate high prediction accuracy with a Mean Relative Error (MRE) below 8.6% within the training sensor domain and robust generalization to external sensors, where the MRE remains below 11.2% across all image types. This work provides a practical and efficient solution for bpp prediction in scenarios where computational resources or time are constrained, facilitating informed compression decisions prior to committing to intensive BPG encoding.