Abstract:Container code is composed of 11 characters

in practice

the 11 code characters can be arranged in many ways

and the distances between neighbor characters in different images vary largely. Traditional methods of automatic container code recognition (ACCR) from visual images are based on isolated character segmentation and isolated character recognition. For those images with code characters being closely arranged

as the characters are difficult to be segmented apart

current methods are difficult to work. To address this problem

a new method

which introduced in HMM based continuous character recognition technique

for ACCR from visual images is proposed. Firstly，in the character segmentation phase

if there are some characters closely arranged

they’ll be extracted as a character region；and for those characters arranged with some distance they’ll be extracted isolatedly. Secondly，in the character recognition phase

for isolated character

a RBF neutral network based isolated character recognition technique is employed

and for closely arranged characters

which are extracted as a string region

HMM based continuous characters recognition technique is employed. Experiment results show that the container code images under various situations can be effectively recognized.