This work proposes a novel adaptive approach for character segmentation and feature vector extraction from seriously degraded images. Experimental results indicate that our algorithm is feasible and effective. These whole properties is put forward for the first time, they are different from those individual property of characters as they are not sensitive to image slant, image stretch and image zoom, and can not be influenced by character ″1″. In the end, all the characters are located and segmented by use of the whole properties of license plate characters in height, centre coordinate and the special proportion relation between the center intervals between two adjacent characters.
Then, a blob analysis is performed to the connected region, from which the middle row passes through. In order to overcome the disadvantage influence bring by plate boundaries, rivets, slant of license plate and uneven brightness of image, firstly, the input image is sharpened by grayscale morphological Top-hat/Bottom-hat transform with a horizontal flat structural element to suppress the background and remove the horizontal boundaries of license plate, which will be advantageous to eliminating character conglutination after the successive OTSU binary threshold.
0 kommentar(er)
