Find Probability that Image Is Text

Question

I can find out whether an image is text:

ImageInstanceQ[img,Entity["Concept", "TextualMatter::m7642"]]

And if an image is most likely text, I can find out the probability that it is text:

ImageIdentify[img, All, 1, "Probability"]

But how do I find the probability that an image is text, even if it's, say, a picture of a lime? I'm looking for the fastest solution possible. My brightest idea at this point is to do a binary search over ImageInstanceQ with different recognition thresholds.

EDIT: I've found a method. Load the net:

imId = NetModel["Wolfram ImageIdentify Net V1"]

And then

imId[image, {"Probability", 
  Entity["Concept", "TextualMatter::m7642"]}]

But this takes on average 0.1 seconds for a 100x100 image. Is there any way to speed this up? I need to compute it about 20000 times.

Answer 1

One method to achieve this is by loading the underlying neural network:

imId = NetModel["Wolfram ImageIdentify Net V1"]

And then

imId[image, {"Probability", 
  Entity["Concept", "TextualMatter::m7642"]}]

Although this is very slow. If anyone has a faster method I'd love to hear it.

Answer 2

Since I've trained a NN classifier in Wolfram Mathematica to identify a text image and get pretty good results.

I tell you that, you can use a lightweight net-structure like SqueezeNet

You can test the inference performance by NetModel["SqueezeNet V1 .1 Trained on ImageNet Competition Data"]

Maybe a LeNet is enough?