The paper presents an improved new method for blurring and uneven light QR code image binarization. Our method is based on local threshold method of Bernsen method. After processing the image with grayscale and Gaussian smoothing filter, it introduces the global threshold based on the variance between classes of Otsu method and the local window gray value mean square error. Different thresholds are selected by multiple rules to achieve binarization of the image. The experimental results show that this method can get more complete images and less noise, and effectively overcomes the effects of blurring and uneven light.