End-to-end text image machine translation (TIMT) aims at translating source language embedded in images into target language without recognizing intermediate texts in images. However, the data scarcity of end-to-end TIMT task limits the translation performance. Existing research explores aligning continuous features from related tasks of text image recogni- tion (TIR) or machine translation (MT) to alleviate the prob- lem of data limitation, but the alignment in continuous vector space is extremely difficult and it inevitably introduces fit- ting errors resulting in significant performance degradation. To better align TIMT features with MT semantic features, we propose a novel Vector Quantization Knowledge Transfer (VQKT) method that employs a trainable codebook to quan- tize continuous features into discrete space. The quantization distribution of the MT feature is utilized as the teacher distri- bution to guide the TIMT model to generate similar discrete codes. Through alignment and knowledge transfer based on probability distribution, the TIMT model can better imitate the feature representation of the MT teacher model and gen- erate high-quality target language translation. Extensive ex- periments demonstrate VQKT significantly outperforms the existing end-to-end TIMT performance.