The paper aims to propose a QR code pixel value reconstruction algorithm based on binary random forest for real life QR codes which are vulnerable to non-uniform illumination and thus cannot be recognized. According to bimodal features and equal gradient value of QR code image, local pixel features of the damaged QR code under non-uniform illumination were extracted, and a single pixel value in the middle of the local QR code matrix was determined with the random forest classification method, so as to gradually realize reconstruction and recovery of all pixel values of the damaged QR code. The experiment showed that, compared with other methods, this algorithm model can extract the real pixel value of QR code by using local features, and recover the damaged QR code image. The experimental results had high level. The QR code reconstruction algorithm based on binary random forest can deal with the problem of non-uniform illumination recognition error, and can be widely applied to the QR code recognition process in life. It has strong practicability.