Disclosed is a virtual viewpoint synthesis method based on image local segmentation, which relates to the digital image processing technology. By mapping the input left and right images to the virtual viewpoints so as to be fused to obtain a synthesized image, smoothing and denoising the rough and noisy depth maps based on the object segmentation information of the scene, the method as disclosed solves the occlusion issue through local area segmentation during the process of viewpoint synthesis, which may guarantee that the subjective quality of viewpoint synthesis will not be significantly deteriorated when the depth map has a relatively large flaw, and maintain geometric information of the scene to the utmost extent so as to generate a real immersive sense, thereby ameliorating the drawback of significant deterioration of synthesis quality in conventional methods when the depth information of the scene has errors and noises, and offering a relatively strong robustness to the errors in the depth map information of the scene. The disclosed method may be applied to a video surveillance system and image processing software, etc.