Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition | Read Paper on Bytez