Abstract:

In order to solve the non-robustness for affine moment invariant values under the change of light conditions, a normalized light affine moment invariants is proposed which is based on the existing illumination model and Jan Flusser affine invariant moment model so as to deduce a normalized light affine invariant moment. The robustness of this proposed method in various kinds of affine transformations is theoretically proved and the invariance for the method is verified by comparing with experimental results in different conditions of light intensity, direction and color. The experimental results show that this method is suitable for grayscale images and color images. Under the condition of light with direction change, the recognition capability of images of the skew movement is improved by 35%. Under the condition of linear light intensity change, the recognition capability of images of the skew movement is improved by 85%. Under the condition of the light with color change, the recognition capability of image of translation and rotation is improved by 7.8%. Under the condition of the light change, taking multiple satellite model images as an example, the recognition capability of image is improved by 36%. This method with simple calculation is suitable for the identification of target spacecraft.

Key words: affine transforms, moment invariant, illumination invariant, object recognition, image recognition