Hybridized KNN-Random Forest Algorithm: Image Demosaicing with Reduced Artifacts

Abstract

Demosaicing is a necessary step in the image processing process in many digital colour cameras. The demosaicing approach creates a full-colour image from a single-sensor array raw image enclosed with a colour filter array. This work proposes a hybrid technique for automatically identifying CFA patterns and demosaicing methods from noise variance distributions. The image interpolation is completed by using the previously demonstrated G, R, and B planes using five techniques, viz. linear, nearest, cubic, rational, v4 for 7 × 7 kernel size. The degree of sharpening to be tested on each image was determined using fundamental experimental findings. The simulation findings show that the KNN and random forest algorithms improve the efficiency of the original images by reducing false colours. Furthermore, the suggested hybrid technique outperforms earlier demosaicing algorithms in terms of average PSNR measurement. Also, the results for structural similarity index and mean structural similarity index justify the significance of reported work.