This study proposes a method of blending visible and near-infrared (NIR) images to enhance their edge details and local contrast based on the Laplacian pyramid and principal component analysis (PCA). The Laplacian pyramid is a multi-resolution decomposition process that separates the edge and base images on each layer. PCA is an image-fusion method that utilizes the distribution of data in an input image. The proposed method implemented both the Laplacian pyramid and PCA to generate a radiance map. The layers of the decomposed visible and NIR images at the same level were synthesized by weights using the PCA algorithm. The calculation method of the PCA algorithm was utilized to adjust the calculation of near-zero-pixel values in an edge-layer image.
Using the PCA algorithm, the soft-mixing method and the mask-skipping filter were applied when the images were fused. The mask-skipping filter skips the PCA algorithm calculation in a specific area, which was proposed given that the edge images of the visible and NIR scenes decomposed by the Laplacian pyramid had a value of nearly zero pixels, which can affect the image fusion. After the mask-skipping filter implementation, image fusion was performed using the soft-mixing method, which selects the transition region using specific pixel values and performs the visible and NIR synthesis of both edge- and base-image fusion. The proposed fusion algorithm utilizes the zero-pixel values in the PCA fusion algorithm. In summary, this study aims to generate a radiance map using the proposed algorithms. Finally, this study proposes a color compensation method using a luminance-channel ratio of the CIELAB color space between the radiance map and the input visible image to preserve the input visible image chrominance in our proposed radiance map.
A camera module was used to simultaneously capture visible and NIR scenes for obtaining several visible and NIR images. The proposed image-fusion method was compared with the conventional method using its reference images and other images captured using the camera module, which comprises two charge-coupled-device (CCD) wideband cameras and a plate beam splitter. The resolution of the CCD camera is 1920 × 1080. The plate beam splitter is a device that splits light, transmits visible light on one side, and reflects the infrared light on the other. Furthermore, a homography algorithm aligned the visible and NIR scenes, given that the CCD cameras had different angles.
The proposed method, with the camera module, was used to obtain visible and NIR scenes. Moreover, the proposed algorithm exhibited better edge details and local contrast compared with conventional visible and NIR image-blending methods.