The invention discloses a vision-based image-text defect detection method, and relates to the field of computer vision recognition. According to the method, a color high-definition industrial camera and a lens are adopted for photographing printed matters within the size, standard templates are manufactured for qualified products, feature information of the templates is extracted, and then printing samples placed in any posture are photographed and compared; according to the algorithm, high-precision multi-scale registration correction is automatically performed on sample images and template images in an image comparison process, so that complex work such as high-precision placement of a single printed matter by using a jig or manual correction of a rotation angle at present is omitted, and deformation generated by the samples in different placement postures can be automatically corrected; and finally, small image-text printing defects can be identified. Not only can characters and symbols be processed, but also various printing contents containing images, characters and forms can be processed, various defects can be better detected, and the high-standard quality detection requirements of the fields of 3C, food, medicine and the like on printed matters are met.