To improve the image scrambling and encryption effect in traditional two-dimensional discrete Arnold transform, a new nonlinear transform for image scrambling is proposed which improves the classical discrete Arnold transform with quasi-affine properties, and can be applied in image scrambling and encryption researching. This method first makes good use of the construction thought in classical discrete standard map, and embeds the nonlinear expressions of output results of one congruence equation for classical two-dimensional discrete Arnold transform into the input item of the other congruence equation for two-dimensional discrete Arnold transform. Then a new transform with good nonlinear characteristics is constructed on the basis of classical two-dimensional discrete Arnold transform in order to quickly improve the scrambling effect of the gray image. In the end, through mathematical proof it is shown that the proposed transform no longer has the quasi-affine invariance properties in the existing two-dimensional discrete Arnold transform, but it is still a reversible mapping with periodic properties; and when it is applied in image scrambling encryption, the original image can be restored from the scrambling and encryption in gray image for its periodic properties or inverse transform. Some experimental results show that the proposed nonlinear transform is effective, and can obtain better scrambling and encryption quality than the existing discrete two-dimensional Arnold transform, meanwhile it is more practical than the standard Arnold transform in view of security.