Digital images are widely used in multimedia technology, making image processing particularly important. Image denoising and segmentation, as fundamental issues in image processing, are crucial for subsequent analysis and understanding. This paper proposes a multi-scale and multi-directional statistical modeling method based on Contourlet Transform. It introduces a Local Context Hidden Markov Model to comprehensively describe the correlation of contourlet coefficients. Additionally, a domain-space adaptive threshold denoising method based on elliptical direction window signal estimation, and an image segmentation technique based on domain models and improved contextual structures, significantly improve the performance of image processing. Furthermore, the feasibility of combining neural network models with contourlet transform is discussed to construct a more efficient and robust image processing framework. The proposed models and methods are validated through simulation experiments, providing new research directions and technical support for future image-processing technologies.