# A3-HongWu

### From CS294-69 Image Manipulation and Computational Photography Fa11

**Name:** Hong Wu

**Assignment: ** Assignment 3

**Course: ** CS294-69, Fall 2011, UC Berkeley

**Paper Implemented:** GradientShop: A Gradient-Domain Optimization Framework for Image and Video Filtering. Pravin Bhat, C. Lawrence Zitnick, Michael Cohen and Brian Curless. (pdf)

## Details:

Our formulation is greatly inspired by the recent work on gradient domain image filtering where direct manipulation and use of image gradients has played a central role. These methods rely on the fact that gradients are integral to the way in which we perceive images.

Base on the fact that Human Vision System (HVS) largely relies on edges and contract, the above paper focuses on easily specifying both zeroth order constraints (i.e., desired pixel values) and first order constraints (i.e., desired pixel gradients in space) in the optimization.

The paper proposes a formula to minimize the energy which contains both data cost and gradient cost.

in which

d provides data constraint for each pixel in desired image f. g is the desired derivative in x and y direction. The image w provides the weight on d and two directions of g.

Instead of using naive weight function

the paper applies a formula based on L1-norm to reduce the influence of the noise.

## Applications:

Figure: Original image of dolphin.

**Smart sharpening**

Based on long edge detecting, the paper only enhances the magnitude of gradients along the long edges rather than along gradient magnitude. By doing this, it avoid to enhance image noise or background clutter.

Figure: Long edge enhancement of dolphin.

**Pseudo image relighting**

The paper add artificial lighting resource by manually specifying a lighting direction. And then, the algorithm boosts the gradients according to the angle between artificial lighting direction and the gradient direction.

Figure: Relighting of dolphin by adding artificial lighting resource at up-right corner.

**Painterly rendering**

The purpose of this application is to photographs into a painterly look. The method is to exaggerate salient features and abstract non-salient features. In another word, the method decrease the gradient at short and no edge and increase the gradient at long edge region.

Figure: Dolphin with painterly rendering effect.

## Other Images:

Figure: Original image of beauty.

Figure: Long edge enhancement of beauty.

Figure: Relighting of beauty by adding artificial lighting resource at up-right corner.

Figure: Beauty with painterly rendering effect.