License plate recognition training data synthesis method

The invention relates to a license plate recognition training data synthesis method. The method comprises the following steps: S1, creating a background image; traversing each pixel in the backgroundimage, wherein each pixel has a random probability p1 to generate different basic element patterns on the background image; S2, randomly selecting a vehicle body image from the vehicle body image database, randomly cutting out a local vehicle body image from the selected vehicle body image, and combining the background image and the local vehicle body image to form a composite image; and S3, randomly selecting an area on the composite image to draw a license plate image. The license plate recognition training data synthesis method has the advantages of being economical, efficient and large inamount. The problems that a traditional manual data collection method consumes manpower and material resources and is low in efficiency are solved. Infinite training data can be quickly provided for alicense plate recognition