Modified DeCanv Method for Quality Improvement using DWT in Digital Images of Paintings

Abstract

Canvas used for painting is obtained by interleaving horizontal and vertical threads in a periodic fashion and is mainly used as a support for paintings. The fabric can be cotton duck, linen, jute, synthetic fiber etc. The cotton duck canvas is the most common type of canvas. Linen canvas has superior quality as the threads are finer and the weave is tighter. Both digital photographs and X-ray images are common investigation tools in art history and art conservation. It can be used for dating or authentication of paintings, for determining the thread counts in art forensics, digital inpainting of cracks etc. It also helps to determine the artist’s style of painting by inspecting the brushstrokes. Digital images from high resolution digital photographs and X-ray can be heavily affected with the underlying canvas structure. The canvas structure present in the paintings can hamper its visual reading by art experts. Therefore, digital removal of canvas helps art conservators to better judge the state of the painting or to determine its history. This paper aims at visual enhancement of the digital image of the painting by removing the canvas components from the frequency domain. The existing method, DeCanv uses Cartoon -Texture decomposition to decompose the input image into a cartoon part and a texture part. Discrete Fourier Transform (DFT) is applied to texture part and is followed by multiscale adaptive thresholding technique to remove the peaks. It is then subtracted from the input image. The proposed method, modified DeCanv enhances the performance by using Discrete Wavelet Transform (DWT). DWT is applied to both input and resultant image and combined in a specific way for better removal of the canvas. The quality improvement is assessed using the parameters Mean Square Error (MSE), Naturalness and Sharpness. This method reduces the MSE and increases the naturalness and sharpness to get a better canvas free image.