Waterjet-assisted laser processing technology provides a new direction in the development of laser processing that can effectively mitigate thermal damage to materials during laser processing; thus, it has good prospect for microfabrication applications. During processing, the quality of the material surface can be inspected in real time, and the introduction of microscopic systems can provide a more convenient observation of material processing details as well as improve the corresponding efficiency. However, because of the interference of the waterjet, bubbles are present in the source image captured by the CCD camera, obscuring detailed information and blurring the surface. In addition, the source image sequence is only partially focused because of the limited depth-of-field of the microscope system. To enhance the detail of surface features of materials for the purpose of dehazing, an image processing technique is employed to improve the contrast of an image by emphasizing the respective brightness, saturation, and textural features of local areas. However, conventional dehazing algorithms begin with the image itself. Without considering that the local area processing method has problems including loss of detailed information and excessive contrast enhancement, development of an image dehazing method for the local areas of different images is necessary. In the traditional microscope mode, the effect of extending the depth-of-field can be achieved via component deformation; however, this has reduced efficiency, large equipment size, and low lateral resolution. Currently, image fusion based on the transform domain is a popular research topic, and is one of the most widely used and mature methods in practical applications.
This study solves the problems of waterjet interference and image fusion. Bubbles were maximally reduced via foreground segmentation and morphological theory, and the detailed information of the source images was enhanced according to the adaptive multiscale Retinex dehazing algorithm. After dividing the image into blocks, the detail index is defined by the standard deviation value of each block. The most suitable Gaussian filter function scale value was determined, and the corresponding weights were calculated to linearly superimpose the single-scale Retinex algorithm of different scales. The source image was decomposed into detail and approximate components using discrete wavelet transform, the detail component was stretched according to the human vision principle, and each discrete wavelet inverse transform was performed according to the fusion rules. As a result, a full-focus image with an extended depth-of-field can be realized.
The experimental system adopted in this work was set up as depicted in Fig. 1, mainly constituting a laser processing and image capture system, and a high-pressure waterjet assist system. The processing and image capture system comprises a laser, waterjet nozzle, and CCD camera. The processed material surface images of the waterjet interference problem are presented in Fig. 5: (a) displays one of the source images captured in the waterjet environment and (b) shows the final processed image. The interference of bubbles on the image information in the processed image appears to fade, and more detailed features are restored. Compared with the source image, the average gradient (AG), standard deviation (SD), and spatial frequency (SF) of the processed image improved by 27.6%, 20.1%, and 4.74%, respectively. Table 3 presents the image quality comparison results of the three source images in the air and waterjet environments. In the waterjet environment, the image quality was significantly lower than that in the air environment, where the SF and AG were reduced by 33.88% and 31.11%, respectively. Figure 6 shows the source images of the material surface collected for waterjet nozzle diameters of 0.4, 0.8, and 1.2 mm as well as the processed images after algorithmic processing. As the diameter of the waterjet nozzle increases, the thicker the flowing water layer on the surface of the sample, and the more limited the image information that can be obtained from the source image. According to Fig. 7(b), the three indicators of the processed image obtained with the diameter nozzle of 0.4 mm reached 95.41%, 71.88%, and 67.29%, respectively. Fig. 8 displays the source images of the material surface collected for waterjet inclination angles of 30°, 45°, and 60°, besides the processed images after algorithmic processing. As the waterjet inclination angle decreases, the thicker the flowing water layer on the surface of the sample, the more limited the image information that can be obtained from the source image. As shown in Fig. 9(b), the three indicators of the processed image considering the 45° angle reached 90.59%,72.69%, and 94.50%, respectively; thus, maximizing the exclusion of the interference of the waterjet, which could restore part of the detailed features.
In waterjet-assisted laser processing, microscopic images of material surfaces are subject to waterjet interference and depth-of-field limitations. Therefore, we propose a waterjet laser processing image-fusion algorithm based on the Retinex dehazing algorithm. The method determines different Gaussian filter function scales for various images, improves the dehazing effect of traditional algorithms, and stretches the detailed components according to human vision. As a result, detail of source image information was enhanced and the image quality was improved. Experiments demonstrate that the algorithm reduces the interference of the waterjet in the source image, enhances the detailed information of the image, and achieves full focus. As the diameter of the waterjet nozzle increases and the inclination angle decreases, the water film on the surface of the material becomes thicker, and the quality of the source image is subsequently reduced, which is difficult for the algorithm to process. However, the experimental results show that with the nozzle diameter of 1.2 mm or an inclination angle of 30°, the processed image still presents most of the detailed features and improved image quality. Thus, the developed algorithm can effectively improve the efficiency of waterjet-assisted laser processing and is expected to be widely used in liquid-assisted laser processing.
