GaussianImage is a novel image representation and compression method that uses 2D Gaussian Splatting to achieve high performance with low GPU memory usage and fast rendering speeds. The method represents images using 2D Gaussians, each defined by position, covariance, color, and opacity. This approach reduces the number of parameters compared to 3D Gaussian Splatting, leading to a 6.5× compression and a 7.375× improvement in compression ratio. A unique rasterization algorithm based on accumulated summation replaces depth-based sorting and alpha blending, enabling efficient rendering without the need for sorting or complex transparency calculations. This results in faster training and inference speeds, with rendering speeds reaching up to 2000 FPS. The method also integrates vector quantization to build an image codec, achieving competitive rate-distortion performance comparable to existing compression-based INRs like COIN and COIN++. Experimental results show that GaussianImage outperforms these methods in terms of compression efficiency and rendering speed, with a fast decoding speed of approximately 2000 FPS. The method is also compatible with partial bits-back coding, further improving compression performance. Overall, GaussianImage offers a compact, efficient, and high-quality image representation and compression solution.GaussianImage is a novel image representation and compression method that uses 2D Gaussian Splatting to achieve high performance with low GPU memory usage and fast rendering speeds. The method represents images using 2D Gaussians, each defined by position, covariance, color, and opacity. This approach reduces the number of parameters compared to 3D Gaussian Splatting, leading to a 6.5× compression and a 7.375× improvement in compression ratio. A unique rasterization algorithm based on accumulated summation replaces depth-based sorting and alpha blending, enabling efficient rendering without the need for sorting or complex transparency calculations. This results in faster training and inference speeds, with rendering speeds reaching up to 2000 FPS. The method also integrates vector quantization to build an image codec, achieving competitive rate-distortion performance comparable to existing compression-based INRs like COIN and COIN++. Experimental results show that GaussianImage outperforms these methods in terms of compression efficiency and rendering speed, with a fast decoding speed of approximately 2000 FPS. The method is also compatible with partial bits-back coding, further improving compression performance. Overall, GaussianImage offers a compact, efficient, and high-quality image representation and compression solution.