What to read after Hierarchical Digital Image Inpainting?

Inpainting is a technique used by artists to fill in the damaged portions of a painting.

They use the colors that are present in the boundary of the damaged portion and

extend them smoothly from the border. Once the major colors are reproduced they

fill in the fine details by looking at the surrounding undamaged portion. This process

is informally called as filling in the gap.

Digital image inpainting is done on digital images to fill in the cracks or to remove

the unwanted text in the picture. When it tries the recover the image hidden by an

object or a person, it is more generally addressed under Disocclusion. When

generating a pixel in the higher resolution, it over laps with Super resolution

algorithms.

Digital image inpainting is generally used in artistic world in recreating the missing

portion or for removing an object to create special effects.

A digital inpainting algorithm does not have knowledge of the missing values, it

utilizes the information present in the known area to fill in the gap. The perfection in

the propagation of the color information (photometry) and propagation of the

structures (geometry) decides the success of the algorithm.

The existing algorithms either fail in the propagation of the photometry for a larger

area and/or fail in the propagation of the geometry. Partial Differential Equation

(PDE) based inpainting algorithms fail in propagating the information for a larger

area and texture synthesis based inpainting algorithms fail in propagating the

structures. Certain algorithms, such as Exemplar based inpainting algorithms,

achieve both the propagation for a limited exten

The existing algorithms either fail in the propagation of the photometry for a larger

area and/or fail in the propagation of the geometry. Partial Differential Equation

(PDE) based inpainting algorithms fail in propagating the information for a larger

area and texture synthesis based inpainting algorithms fail in propagating the

structures. Certain algorithms, such as Exemplar based inpainting algorithms,

achieve both the propagation for a limited extent.

The technique presented in this thesis is inspired by the response of Human visual

system in viewing a photograph from various distances. When viewed from a farther

distance, the coarser information in the image could only be viewed. As the distance

reduces, the finer details could be seen. Hence filling in the coarser information and

then the finer information will result in better reconstruction. The Phenomenon is

best imitated by the hierarchical multi resolution models. This model is used in

Hierarchical Total Variation (TV) inpainting, Hierarchical exemplar based

inpainting, Hierarchical Wavelet based inpainting and Hierarchical Laplacian based

inpainting.

Existing PDE based TV algorithm and Exemplar algorithm are modified in

Hierarchical TV inpainting and hierarchical Exemplar based inpainting respectively

to suit the hierarchical structure. Two different variation of Exemplar based

inpainting viz. Hierarchical search space refinement and Hierarchical filling

algorithms are suggested under Hierarchical Exemplar based inpainting.

When few interior values are known then the filling-in becomes easier. This

criterion is achieved by the Top down approach of filling in the hierarchical model.

This method is followed in all the hierarchical models except Hierarchical search

space refinement algorithm.