Intrinsic image modulation characters extraction based on monogenic and strcture tensor

Image modulation represents image by meaningful characters such as image instantaneous amplitude and instantaneous frequency. A new image modulation method is proposed in this work. In detail, the Bidimensional Empirical Mode Decomposition (BEMD) is improved to adaptively decompose an image into mono components. Then, the dominant orientation of image features in a local neighborhood is estimated by a more robust structure tensor-based method. In addition, using the monogenic signal in the preferred orientation, the key Amplitude Modulation and Frequency Modulation (AM/FM) parameters including the instantaneous amplitude, the instantaneous phase and the instantaneous frequency are extracted. We illustrate proposed techniques on natural images, and illustrate its feature-extraction capabilities.