Pattern recognition methods should be robust against small changes, yet be insensitive to certain modifications such as changes of position, intensity, orientation, scale and color. Those two requirements are somewhat contradictory. For example, linear filtering methods are invariant to changes in position, but by definition changes in the outputs are proportional to changes of intensity of the objects. We shall discuss recent advances in invariant pattern recognition, including nonlinear methods that allow maintaining the shift invariance of linear filtering while being insensitive to changes of intensity of objects of interest. Detection of targets in camouflage and multiobject recognition will also be discussed.