Film grain reduction on colour images using undecimated wavelet transform

The presence of film grain often imposes the crucial quality choice between film enlargement and speed. In this work we present an automatic technique for reducing the amount of grain on film images. The technique reduces the noise by thresholding the wavelet components of the image with parameterised family of functions obtained with an initial training on a set of images. The training produces the parameters identifying the functions by optimising a cost function related to the image visual quality. The method has been tested on images contaminated by artificial and by real grain noise from two Kodak film makes. Being the main focus of this work on the grain reduction aspect rather than on the modelling side, we rely on a well known and state of the art software (Furnace) instead of producing a new noise model. The results demonstrate the efficiency of the method in reducing the grain noise and the ability of the technique in adapting the parameters to the noise level on each colour component. Another relevant characteristic of the method is its potential to be used for various different applications, class of images and type of noises just by modifying training set of images, cost function and shape of the thresholding functions.