For three-dimensional (3D) reconstruction of targets, existing image-processing algorithms require a single background, and they significantly depend on the experimental environment. Therefore, an image-preprocessing algorithm for 3D reconstruction of targets in complex backgrounds is proposed. First, to maximize the target detail information, Gaussian filtering, Gamma conversion, and histogram-equalization processing are performed on the acquired images to remove image noise and suppress complex backgrounds. Then, the Grab cut and Deeplab algorithms are combined to solve the problems of long time consumed on Grab cut and edges blurred on Deeplab, effectively separating the target from complex backgrounds. A test platform for the car model is built and sixteen sets of target images are obtained to verify the algorithm. Considering two sets of targets as examples, the effects of the proposed algorithm and the traditional 3D-reconstruction image-preprocessing algorithm are compared. The segmentation accuracy of the proposed algorithm is 0.9986, the sensitivity is 0.9889, and the specificity is 0.9991, which are higher than those of the traditional algorithm. The point-cloud noise rate of the traditional algorithm is 22.7%, which is reduced to 1.15% by the proposed algorithm. The average reconstruction time of the proposed algorithm is 2.245 s, which is 60.6% of the time of the traditional algorithm. These results prove that the proposed image-preprocessing algorithm offers superior 3D image reconstruction under complex backgrounds.