We present a new method to detect the dim changing target for the image sequence of outside stereo scene obtained from the airborne camera. In order to compensate for the movement of the platform, a cascading image registration mode is proposed, which firstly uses the biological vision method to register the images to be detected coarsely and fast, and which then uses least square match based on invariant coefficient to register the images accurately. In order to compensate for the gray-scale changes between the images, a new histogram similarity transformation is proposed, which is ?able to handle not only linear gray-scale changes but also nonlinear gray-scale changes. In order to reduce the stereo effect, a sub-region division method is proposed, which divides an image into several sub-regions, and then calculates the affine transformation parameters for each corresponding sub-regions independently. Using this scheme, we can make sure that different sub-regions correspond to different geometric transformation parameters. Experimental results using many real images obtained from electro-optic pods clearly demonstrate the correctness and validity of the method.