This paper presents an adaptive stereo matching algorithm that dynamically selects window sizes and shapes based on local intensity and disparity variations to improve disparity estimation accuracy. The algorithm uses a statistical model to represent uncertainty in disparity estimates within a window, assuming that uncertainty increases with distance from the center of the window. This allows the algorithm to compute both disparity estimates and their associated uncertainties, enabling the selection of windows that minimize uncertainty for each pixel. The algorithm is tested on both synthetic and real stereo images, demonstrating its effectiveness in producing accurate disparity maps. The results show that the adaptive window approach outperforms fixed-size window methods, particularly in handling varying surface textures and disparity edges. The algorithm's ability to adaptively adjust window size and shape leads to more reliable disparity estimates, reducing both random and systematic errors. The method is also shown to produce high-quality 3D reconstructions, as demonstrated by perspective views of recovered scenes and isometric plots of disparity maps. The algorithm's performance is validated on real-world data, including stereo images of a town model and a coal mine, where it achieves improved depth estimation compared to previous methods. The key contribution of the paper is the development of an adaptive stereo matching algorithm that accounts for both intensity and disparity variations within a window, leading to more accurate and reliable disparity estimates.This paper presents an adaptive stereo matching algorithm that dynamically selects window sizes and shapes based on local intensity and disparity variations to improve disparity estimation accuracy. The algorithm uses a statistical model to represent uncertainty in disparity estimates within a window, assuming that uncertainty increases with distance from the center of the window. This allows the algorithm to compute both disparity estimates and their associated uncertainties, enabling the selection of windows that minimize uncertainty for each pixel. The algorithm is tested on both synthetic and real stereo images, demonstrating its effectiveness in producing accurate disparity maps. The results show that the adaptive window approach outperforms fixed-size window methods, particularly in handling varying surface textures and disparity edges. The algorithm's ability to adaptively adjust window size and shape leads to more reliable disparity estimates, reducing both random and systematic errors. The method is also shown to produce high-quality 3D reconstructions, as demonstrated by perspective views of recovered scenes and isometric plots of disparity maps. The algorithm's performance is validated on real-world data, including stereo images of a town model and a coal mine, where it achieves improved depth estimation compared to previous methods. The key contribution of the paper is the development of an adaptive stereo matching algorithm that accounts for both intensity and disparity variations within a window, leading to more accurate and reliable disparity estimates.