Section: New Results

Computational Photography

Participants : Cyril Soler, Kartic Subr, Adrien Bousseau, Nicolas Holzschuch.

Empirical Mode Decomposition of Images based on Local Extrema

Participants : Kartic Subr, Cyril Soler.

Inspired by work on Huang et al. on empirical mode decomposition of 1D signals, we have designed an equivalent method for decomposing images into layers of details. Our method has a very unique advantage over existing algorithms of layer decomposition: the notion of details we use does not depend on contrast but rather on image-space frequency. Besides, our decomposition preserves edges, even when their amplitude is lower than the amplitude of surrounding details.

To achieve this, we base our decomposition algorithm on the extraction of an upper and a lower envelop image which respectively interpolate local maxima and minima. Each interpolation is made edge-preserving by examining the local variance between maxima (resp. minima). We then average the two envelops to obtain an edge-preserving smoothing of the input image. The detail layer is extracted by removing the smoothed image from the input image. The algorithm can be performed hierarchically by feeding the smoothed layer to the decomposition algorithm again (Figure 13 ).

Our decomposition method allows to perform a considerable number of interesting applications, including from multi-scale contrast enhancement, noise and detail removal, tone mapping and hatch-to-tone filtering. This method has beed published to Siggaph Asia 2009 [19] .

Figure 13. Empirical mode composition applied to a image where multiple levels of high contrast detail exist. Right: The different modes that our method automatically extracted sensibly correspond to these levels of detail.

User-Assisted Intrinsic Images

Participant : Adrien Bousseau.

In the context of an intership at MIT, Adrien Bousseau has work with Frédo Durant and Sylvain Paris on the decomposition of a photograph into the product of an illumination component that represents lighting effects and a reflectance component that is the color of the observed material. This is an under-constrained problem and automatic methods are challenged by complex natural images.

In this work, we describe a new approach that enables users to guide an optimization with simple indications such as regions of constant reflectance or illumination. Based on a simple assumption on local reflectance distributions, we derive a new propagation energy that enables a closed form solution using linear least-squares. We achieve fast performance by introducing a novel downsampling that preserves local color distributions. We demonstrate intrinsic image decomposition on a variety of images and show applications.

This paper has been publish at Siggrapg Asia 2009 [12] .