According to the optical characteristics,defect formation mechamism and characte ristics of the light guide plate,this paper proposes a method of the defect dete ction based on machine vision.Firstly,according to the distribution characteri stics of the light guide points in the light guide plate,an automatic density-based partitioning algorithm is proposed.Secondly,a Gaussia n bidirectional guide filter is respectively designed to convolve the image of the light guide plate for differe nt partitions.Differences between the different sizes of the average filter are used to remove the interference from t he normal light guide points to obtain the defect-enhanced image.Then,the binarized threshold operation is used to e xtract the connected regions of the suspected defects.The final result is based on the analysis of its regional cha racteristics,which will be classified. Finally,a large number of experiments were carried out on the basis of the ligh t guide plate images collected in the industrial field.The experimental results show that the detection accuracy is hi gh for bright spot,compressi on and line scratch defects.It can basically meet the requirements of industrial detection.