This paper presents an adaptive window-based stereo matching algorithm that addresses the challenge of selecting an appropriate window size for computing correlation or sum of squared differences (SSD). The algorithm evaluates the local variation of intensity and disparity to select windows that minimize the uncertainty of disparity estimates. A statistical model is employed to represent the uncertainty of disparity, which increases with the distance from the center point. This model allows the algorithm to compute both the disparity estimate and its uncertainty, enabling the selection of optimal windows for each pixel. The method controls both the size and shape of the window. Experimental results on synthetic and real images demonstrate the effectiveness of the algorithm, showing improved performance over fixed-size window methods in terms of disparity accuracy and noise reduction.This paper presents an adaptive window-based stereo matching algorithm that addresses the challenge of selecting an appropriate window size for computing correlation or sum of squared differences (SSD). The algorithm evaluates the local variation of intensity and disparity to select windows that minimize the uncertainty of disparity estimates. A statistical model is employed to represent the uncertainty of disparity, which increases with the distance from the center point. This model allows the algorithm to compute both the disparity estimate and its uncertainty, enabling the selection of optimal windows for each pixel. The method controls both the size and shape of the window. Experimental results on synthetic and real images demonstrate the effectiveness of the algorithm, showing improved performance over fixed-size window methods in terms of disparity accuracy and noise reduction.