Illumination normalization for image restoration using modified retinex algorithm

Abstract

This paper presents a novel method of illumination normalization based image restoration. A modified retinex algorithm is proposed to remove the shadow and restore the image. First, image is split into illumination (L) and reflectance (R) components. The Reflectance component is subjected to threshold filtering while the illumination component is subjected to modified retinex algorithm and the resulting reflectance component is combined effectively with the output of threshold filter for obtaining the shadow-free image. Illumination normalization is performed on both small-scale as well as large-scale features. Using this approach, face images with cast shadows are normalized efficiently. The quality of the illumination normalized image is evaluated by means of JPEG quality score and PSNR values. We observed very good quality score for illumination normalized images in comparison with original images. The proposed method has a great potential in real-time face recognition systems, especially under harsh illumination conditions.