The present disclosure discloses a method and device for localized blind watermark generation and detection. The method for localized blind watermark generation may include: generating an N-dimensional pseudorandom initial sequence according to a preset key, and generating a 2N-dimensional sequence by inverting the initial sequence bit by bit and appending the inverted sequence to the initial sequence; and extracting the first 2N coefficients of a feature point of an image to form a coefficient sequence, and acquiring a parity of a bit of the coefficient sequence according to a parity of a corresponding bit of the 2N-dimensional sequence. With the present disclosure, redundant expansion may be performed on watermark data to increase the scale of test data with respect to the original bit data before embedding the test data into a transformation domain of the vicinity of a feature point of an image; during detection, a sequence extracted from the transformation domain of the vicinity of a certain feature point is checked bit by bit, and it is determined whether a watermark is embedded in the vicinity of the point base on the result of the examination, thereby enhancing a watermark detecting efficiency.