Abstract: It is prerequisite to accurately extract the character of molten pool image for analyzing the welding quality.Frequency method of image processing has more advantages compared with the traditional spatial methods.The disadvantage of the later is sensitive to noises which can be overcome by the new method.The image character can be sharpened by image processing frequency methods. First, the Median-Winner filtering was used to eliminate the noises of the molten pool images.Then, the Butterworth high-pass filter based on the Fourier transform was used to strengthen the molten pool image. Also, the high frequency enhancement filtering arithmetic was applied to compensate the molten pool images.The molten pool images were processed by histogram equilibration.Finally, the Canny edge detection algorithm and erosion operation of mathematical morphology were applied to detect the molten pool edge.The experimental results showed that the improved high frequency enhancement method combined with the image processing spatial methods were effective to extract the molten pool image character.