Number recognition from Image with python and tensorflow

Question

Details: Ubuntu 14.04(LTS), OpenCV 2.4.13, Spyder 2.3.9(Python 2.7), Tensorflow r0.10

I want to recognize Number from the image with Python and Tensorflow(optional OpenCV).

Additionally, I want to use MNIST data training with tensorflow

Like this(the code is referred to this page's video),

Code:

import tensorflow as tf
import random

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

x = tf.placeholder("float", [None, 784])
y = tf.placeholder("float", [None, 10])

W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))

learning_rate = 0.01
training_epochs = 25
batch_size = 100
display_step = 1

### modeling ###

activation = tf.nn.softmax(tf.matmul(x, W) + b)

cross_entropy = tf.reduce_mean(-tf.reduce_sum(y * tf.log(activation), reduction_indices=1))

optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cross_entropy)

init = tf.initialize_all_variables()

sess = tf.Session()
sess.run(init)

### training ###

for epoch in range(training_epochs) :

    avg_cost = 0
    total_batch = int(mnist.train.num_examples/batch_size)

    for i in range(total_batch) :

        batch_xs, batch_ys =mnist.train.next_batch(batch_size)
        sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys})
        avg_cost += sess.run(cross_entropy, feed_dict = {x: batch_xs, y: batch_ys}) / total_batch

    if epoch % display_step == 0 :
        print "Epoch : ", "%04d" % (epoch+1), "cost=", "{:.9f}".format(avg_cost)

print "Optimization Finished"

### predict number ###

r = random.randint(0, mnist.test.num_examples - 1)
print "Prediction: ", sess.run(tf.argmax(activation,1), {x: mnist.test.images[r:r+1]})
print "Correct Answer: ", sess.run(tf.argmax(mnist.test.labels[r:r+1], 1))

But, the problem is how can I make numpy array like

Code addition:

mnist.test.images[r:r+1]

[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 0.50196081 0.50196081 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 1. 1. 1. 1. 0.50196081 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 1. 1. 1. 0.50196081 0.50196081 0.50196081 0.74901962 1. 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 0.74901962 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 1. 1. 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 0.74901962 1. 1. 1. 1. 0.74901962 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 1. 1. 0.74901962 0. 0. 0. 0. 0. 0.25098041 0.50196081 1. 1. 1. 1. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 1. 1. 1. 0.50196081 0.50196081 0.74901962 1. 1. 1. 1. 1. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.74901962 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.25098041 1. 1. 1. 1. 1. 1. 1. 0.50196081 0.25098041 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.50196081 0.50196081 0.50196081 0.50196081 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. ]]

When I use OpenCV to solve the probelm, I can make numpy array about the Image but little bit strange. (I want to make array into a vector of 28x28)

Code addition:

image = cv2.imread("img_easy.jpg")
resized_image = cv2.resize(image, (28, 28))

[[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]

[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]

[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]

...,

[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]

[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]

[[255 255 255] [255 255 255] [255 255 255] ..., [255 255 255] [255 255 255] [255 255 255]]]

And then, I put the value('resized_image') into the Tensorflow code. Like this,

Code modification:

### predict number ###

print "Prediction: ", sess.run(tf.argmax(activation,1), {x: resized_image})
print "Correct Answer: 9"

As a result, the error is occured at this line.

ValueError: Cannot feed value of shape (28, 28, 3) for Tensor u'Placeholder_2:0', which has shape '(?, 784)'

Finally,

1) I want to know how can I make a data which can be input the tensorflow code(maybe numpy array [784])

2) Do you know about the number recognition examples that use tensorflow?

I'm a beginner in machine-learning.

Please Tell me in detail what should I do.

Answer 1

It seems like the image you are using is RGB hence the 3rd dimension (28,28,3).

Where as the original MNIST images are greyscale with width and height of 28. This is why the shape of your x placeholder is [None, 784] because 28*28= 784.

CV2 is reading the image in as RGB and you want it to be greyscale ie (28,28) When doing your imread you may find it helpful to use this.

image = cv2.imread("img_easy.jpg", cv2.CV_LOAD_IMAGE_GRAYSCALE)

By doing this you image should have the correct shape (28, 28).

Also CV2 image values are not in the same range as the MNIST images that are being shown in your question. You may have to normalize the values in the image so that they are in the range 0-1.

Also you might want to use a CNN (slightly more advanced but should give better results) for this. See the tutorial on this page https://www.tensorflow.org/tutorials/ for more details.