Successful prediction of epitope-T cell receptor (TCR) interactions can help with effective vaccination and personalized healthcare. Unseen epitope-TCR interaction prediction is based on independent sets of training and testing data, which is good to find their corresponding epitopes for novel, unseen diseases. In this study, we present a framework for predicting the unseen epitope-TCR interactions based on physicochemical properties of constituent amino acids of epitope and TCR CDR3 sequences. Sequence based models for epitope-TCR interaction prediction generally extract features individually from each sequence and then combine them together. However, in this study, the features for the unseen epitope-TCR interaction model have been generated as images from both sequences simultaneously by computing the absolute difference and outer product of two vectors consisting of the physicochemical property values of amino acids. The performances based on nine different physicochemical properties of amino acids have been compared and the best performing properties are selected. Some properties are combined together to achieve the highest performance. The model exhibits much higher performance in comparison with the existing unseen epitope prediction models. The model produces the AUC of 0.64 for absolute difference based features with only two best performing properties, and the AUC of 0.60 for vector outer product with the same two properties. Furthermore, our model achieves the AUC of 0.82 by combining both types of features while the best existing model achieves the AUC of only 0.55 in the setting of unseen epitope-TCR interaction prediction.