Abstract: In the welding process, the welding region has large brightness range and random variation, which result in great disturbance of the weld image and are difficult to extract features. In this paper, a method of obtaining clear image for welding area based on visual attention is proposed by using the directional characteristics of laser fringes and the brightness characteristics of structured light and molten pool in the image. Firstly a unidirectional feature map of structured light fringes under different detection directions is obtained for the weld area image, and the Multi-orientation feature map is obtained by weighted fusion. Secondly, the original image is enhanced by the high-pass Butterworth filter to obtain the intensity feature map. Then, an improved saliency measure method is proposed to realize the good segmentation between the salient region and the complex background image region. Finally, a clear image of structured light and molten pool is obtained by image fusion normalization, and the effect of image saliency detection is evaluated objectively by clustering center distance. The experimental results show that the proposed method is effective.