Predspracovanie obrazu pre optické rozpoznávanie ... - TUKE

More documents

Recommendations

Info

$C:\tst\data_bi.v 1 `timescale 10ns/1ns 2 3 module data_bi(aDATA ...$

FEI TU v Košiciach Diplomová práca List č. 84 Obr. 45: Image after smoothing (mid) and thresholding (right). On the picture 42 on the left is what you will get if you will scan the printed text. Note that the scanned image contains levels of gray that were never printed and that were never designed on the screen of the computer. Image like this is not thresholdable - the threshold is useless and the OCR can’t be used. If you would apply a light smoothing couple times on the image on the picture 44 or a heavy smoothing once, you can get a result like this one: As you can see, the smoothing (in some sort) converted the different levels of gray into less levels of more similiar levels of gray. On the image on the picture 45 in the middle a threshold can be applied with better results than on the image on the left. But as you can see, with threshold you will loose some parts of the text (picture 45 on the right) or you will fail to remove some parts of the background (but the text will not loose anything). So this is the case when the different method of <strong>pre</strong>processing must be used. Design and implementation Most of the standart and well known methods works by determining the new value of the level of shade of the image pixel by multiplying the current level of shade of the pixel and its neighbours by some coeficients. These coeficients are described in a convolution matrix.This matrix can be displayed as a set of input pixels connected with the output pixel with lines with wieghts. (See the picture 46) In artificial inteligence, the most similiar thing to this filter is a forward-feed neural network, because: - the input can be organized in the mask of N × N
FEI TU v Košiciach Diplomová práca List č. 85 in in in 1,1 1,2 1,3 in in in 2,1 2,2 23 in in in 3,1 3,2 3,3 k 2,1 k Obr. 46: Schematical drawing of a 2D image filter. in1,1 in1,2 in1,3 in2,1 in2,2 in2,3 in3,1 in3,2 in3,3 1,3 k 3,1 h 1 h 2 h 3 out out1 out2 Obr. 47: Schematical drawing of neural network - the output can be one (or more) pixels - the signal flows from the input to the output - the input and output are connected with lines which have weights So a very similiar forward-feed neural network to the 2D filter can look like one shown on the picture 47. I’ve chosen the parameters of used neural network: - type of neural network: forward-feed full-connection neural network - learning method: standart error back-propgation - input layer of N × N neurons - hidden layer of M neurons - output layer of 2 neurons (one for direct level of shade and one for inverted level of shade)
Page 1 and 2:
Technická univerzita v Košiciach
Page 3 and 4:
Poďakovanie Chcem poďakovať môj
Page 5 and 6:
Thesis title: Image preprocessing f
Page 7 and 8:
Obsah Formulácia úlohy 1 1 Úvod
Page 9 and 10:
FEI TU v Košiciach Diplomová prá
Page 11 and 12:
Page 13 and 14:
Page 15 and 16:
Page 17 and 18:
Page 19 and 20:
Page 21 and 22:
Page 23 and 24:
Page 25 and 26:
Page 27 and 28:
Page 29 and 30:
Page 31 and 32:
Page 33 and 34:
Page 35 and 36:
Page 37 and 38:
Page 39 and 40:
Page 41 and 42: FEI TU v Košiciach Diplomová prá
Page 91: FEI TU v Košiciach Diplomová prá
show all

Predspracovanie obrazu pre optické rozpoznávanie ... - TUKE

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?