Some common ideas in the [CLPR] localization algorithm

I. INTRODUCTION

How do I find the license plate from a picture? This is an application of machine vision. Naturally, the angle of thinking is from the license plate itself information to start, for discussion convenience, the following are the long narrow blue-white license plate for example.

This is the basic information of a license plate.

At the glance of the past, the information available is: aspect ratio-3.14, character number-7, first character is Chinese, second character is letter, followed by 5 letter/number mixed equidistant arrangement.

It can also be broadly understood that a clear license plate should have enough edge information, in other words, the edge information is dense enough to gather in a 3.14:1 rectangle.

So today's algorithm, all using the edge of information, combined with the morphology to distinguish.

Two. What is Edge?

What is the edge? In general, we define the transition point of grayscale in grayscale (256 order) images as edges. Jump point refers to the similarity of two pixels of the gray difference is large.

The usual method for computing edges is to manipulate images using operators, and we use Sobel operators here. The third-order Sobel operator for detecting vertical edges is roughly the following form:

We set the operator as S, the part of the image applied to the operator is GS, the output is the OS. We can get OS = GS * S. Also calculates the difference between the GS left and right gray values, assigning this difference to the middle three pixels. As you can see, this is in line with our definition of edge. Let's look at the results of the Sobel operator using both vertical and horizontal two directions:

The first one is the original, and the second is the grayscale edge.

Three. Morphology

Four. Summary of common algorithms for license plate recognition in China

Five. Analysis of advantages and disadvantages of edge location algorithm

Some common ideas in the [CLPR] localization algorithm