Image enhancement with wavelet-optimized whitening

TL;DR

This paper introduces a wavelet-based image enhancement method for solar corona images that automatically produces sharp, contrasted results without manual parameter tuning, effectively suppressing noise and halos.

Contribution

It proposes a novel wavelet whitening technique with bilateral modification that enhances solar images objectively and efficiently, outperforming existing methods in speed and quality.

Findings

Produces sharp, contrasted images without manual parameter adjustment

Prevents high-frequency noise amplification through built-in denoising

Outperforms multiscale Gaussian normalization in speed and quality

Abstract

Due to its physical nature, the solar corona exhibits large spatial variations of intensity that make it difficult to simultaneously visualize the features present at all levels and scales. Many general-purpose and specialized filters have been proposed to enhance coronal images. However, most of them require the ad hoc tweaking of parameters to produce subjectively good results. Our aim was to develop a general purpose image enhancement technique that would produce equally good results, but based on an objective criterion. The underlying principle of the method is the equalization, or whitening, of power in the {\it \`a trous} wavelet spectrum of the input image at all scales and locations. An edge-avoiding modification of the {\it \`a trous} transform that uses bilateral weighting by the local variance in the wavelet planes is used to suppress the undesirable halos otherwise produced by discontinuities in the data. Results are presented for a variety of extreme ultraviolet (EUV) and white light images of the solar corona. The proposed filter produces sharp and contrasted output, without requiring the manual adjustment of parameters. Furthermore, the built-in denoising scheme prevents the explosion of high-frequency noise typical of other enhancement methods, without smoothing statistically significant small-scale features. The standard version of the algorithm is about two times faster than the widely used multiscale Gaussian normalization (MGN). The bilateral version is slower, but provides significantly better results in the presence of spikes or edges. Comparisons with other methods suggest that the whitening principle may correspond to the subjective criterion of most users when adjusting free parameters.