The work aims to propose an anti-overflow watermark scheme based on region of interest (ROI), regarding the problem of pixel overflow in watermark embedding in images with a mass of bordered pixels like drawing of engineering design and medical images. The scheme used the improved region growing algorithm to segment irregularly shaped ROI from the original image, then filled the region with pixels to get rectangular region of watermark embedding (ROW). After applying lifting wavelet transform and Arnold transformation to ROW, the scheme partitioned and scanned the region to embed the watermark which was in QR code into the host image with the means of parity mapping, while bordered pixels were skipped. The edge of images with embedded watermark still remained integrated and the appearance of pixel overflow was greatly reduced. The PSNR of watermark embedded images were greater than 48 db, which indicated that the invisibility of watermark was greater than general watermark embedding schemes and satisfied the requirement of watermark invisibility of high-precision images; moreover, after the attacks, the extraction of watermark could still be achieved. The proposed scheme takes full account of the feature of images with large area background and mass bordered pixels, and makes sure that the certified watermark is embedded into critical ROI of the image. The scheme realizes the anti-overflow watermark embedding process on the basis of quality guarantee of the original image. The extracted watermark from images under attack can still show authentication information. The proposed scheme has practical application value in packaging and printing and engineering.