I have a multilayer perceptron model that is trained to recognize handwritten English letters from an image. In the training set each image matrix had 784 pixel values. The labels of these images corresponded to the place in the alphabet. For example: A is 1, B is 2 and so on. In total there were 26 classes. From the pixel values and the corresponding label the model is now capable to predict a letter.

Now I am trying to predict corrupted images with the model. Each corrupted image has 4 letters in it. Each corrupted image has a dimension of (30, 140). I have tried to come up with a solution, but I am at a loss at this point. So let's say I have a corrupted image with pixel values that represent the letters VRLX, I want to predict with my existing model that the label is 22181224 (v = 22, r = 18, l=12, x=24). The current model is capable to predict a letter independently. However, I fail to fit my model with these corrupted images. Does anyone have some suggestions that can send me in the right direction?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.