A Novel Detail Enhancement Method for Industrial Digital Radiography Based on Multiscale Pixel-Level Adaptive Fusion

Abstract

To address the limitations of current methods, such as detail loss, difficulties in enhancing complex features, reliance on powerful computing systems, and difficulties in engineering applications, a novel detail enhancement method for X-ray images in industrial digital radiography is proposed based on multiscale pixel-level adaptive fusion in logarithmic space. In the proposed method, Gaussian convolution is used to construct a multiscale space of the X-ray image. Based on the fact that Gaussian blur and detail enhancement exhibit inverse relationships, the pixel importance for detail enhancement is assessed by the difference between the Gaussian convolved image and the original image. The tanh function and pixel importance for detail enhancement are employed to infer the pixel fusion coefficient, and an enhancement method for X-ray images is achieved through pixel-level adaptive fusion across scales in logarithmic space. The experimental results show that in terms of PSNR, the proposed method improves HE, CLAHE, LCR, DWT, TC-U-NET, and CNN-DEMD by an average of 50.98%, 43.78%, 28.38%, 26.52%, 7.93%, and 5.50%, respectively. The proposed method improves HE, CLAHE, LCR, DWT, TC-U-NET, and CNN-DEMD by an average of 54.68%, 41.40%, 21.95%, 18.65%, 7.04%, and 4.58%, respectively. In terms of SF, the proposed method increases HE, CLAHE, LCR, DWT, TC-U-NET, and CNN-DEMD by an average of 44.48%, 33.25%, 19.56%, 20.13%, 6.17%, and 4.85%, respectively. The experimental findings demonstrate that the proposed method achieves favorable results and exhibits excellent performance.