Image Aesthetics

Automatic assessing image aesthetics is an active topics in computational photography. Our approach is threefold:

1. rather than assessing aesthetics of an image independently of the image collection of which it belongs, we take into account image context (image collection, user preference) while aesthetics assessment
2. image aesthetics assessment just ranks or score an image, we propose methods to analyse image aesthetics. This enables to develop aesthetics aware methods that preserve image aesthetics while media retargeting such as tone retargeting, color transfer, etc.
3. we also propose computational distances to assess how close an image is aesthetically to another image. We use these metrics to automatically light source setup in global illumination

High Dynamic Range Imaging

With HDR imaging come two new technologies: Tone Mapping: compression of the dynamic range of HDR content to SDR  and Tone Expansion: upscaling the dynamic of a SDR content to HDR.
We are involved in both topics. Our main goals are to:

1. preserve artistic intent of original HDR/SDR media when compressed/upscaled to SDR/HDR
2. understand and model brightness perception and comfort
3. propose new approaches to HDR capture

Our major contributions in these domains are:

1. We put into evidence that tone mapping independently each image of a HDR sequences introduces  Temporal Artifacts in the resulting SDR sequence. We proposed a classification of these artifacts
2. We proposed a video Tone Mapping Operator,  called Zonal Brightness Coherency, that reduces/prevents these Temporal Artifacts and preserves the aesthetic intent portrayed by the HDR Sequence
3. We developed a video Tone Expansion operator (vTEo), called Style Aware Tone Expansion, that upscales SDR contents to get HDR once, while preserving the artistic intent. Our vTEo outperforms state-of-the-art vTEo especially when SDR content is of high or low key style. Our Style Aware Tone Expansion won the NAB 2017 innovation award
4. We proposed a new feature (based on statistical measure and user studies) to predict Brightness Comfort. This feature helps prevent from visual discomfort that can occur while looking at too bright contents on HDR displays

Image Editing

We investigate:

1. distribution based (multivariate generalized Gaussian distribution, Beta distribution) method to global image editing such as color transfer
2. new kind of filtering such as bi-local CAT-based guidance filter to tackle many many image editing tasks/processing