Home
Forum: "Grab";
Current archive: 2002.01.08;

Down

#### help someone can Find similar branches

Новенький   (2001-11-07 01:11) [0]

It is necessary to write a laboratory work on pattern recognition. For example, write something in the Times New Romans bmp file and, through a program, output what was written to a text file.

panov   (2001-11-07 02:10) [1]

In my opinion, something like the sample is introduced, then the “mismatch” -fuzziness tolerance is made, and a bit is determined from the bitmap as far as the drawing is close to the original ...

Merlin   (2001-11-07 03:54) [2]

If the font is known in advance and the picture is free of defects, then it still hasn’t gone. But if you write on laboratory FineReader, then I would like to look at your term papers;)

Наталья   (2001-11-07 10:24) [3]

The easiest way.
1. You draw an image on paper in a cell, for example, the letter "A", and counting how many times each horizontal line crosses the image, you get a sequence like "1,2,2,1,2,2", you discard the repeating numbers, you get "1,2,1,2". You do the same thing for vertical lines, get "1,2,1". T.O. the image of the letter "A" can be represented in two sets: horizontally, "1,2,1,2", and vertically, "1,2,1".
2. You determine the "horizontal" and "vertical" sequences of image pixels in a bmp file and compare with a certain early sequence, if they match, then with a high probability, the letter "A" is drawn.
2 paragraph I have greatly simplified, but I think it’s clear which way to go :)

zero   (2001-11-07 15:40) [4]

Last year I wrote such a lab.
Natalya is right.
I just didn’t explain that before paragraph 1 you need to process the image of the letter so that the thickness of the lines is equal to one pixel (this is also not a small job).

Merlin you're wrong, with knowledge of the theory everything is feasible.

zero   (2001-11-07 15:45) [5]

I would like to hear the answer of the respected moderator, why this topic should be in this section and not mainly. Basically, there are a lot of such topics and no one throws them away.

модератор   (2001-11-08 07:40) [6]

Other forums discuss programming on Delphi.

p@s   (2001-11-08 09:41) [7]

Five or six years ago, I read an article in a magazine about pattern recognition. There, even the algorithms were not a bad mathematical base. The magazine was HardAndSoft Toli, Monitor Toli. Sorry for a long time, I don’t remember exactly.

zero   (2001-11-08 10:35) [8]

To moderator
> Other forums discuss Delphi programming.
So a person in Delphi is going to program.
Maybe someone will offer components or something else Delphic.
So you're wrong.

модератор   (2001-11-08 10:42) [9]

> zero
> So a man is going to program on Delphi.
Who said that?

Labert   (2001-11-08 10:55) [10]

It is on such tasks that the difference between programmers and "encoders" is manifested.
Here you can’t do with just one knowledge of the syntax of calling functions. On this issue there is a cloud of literature and wise mathematical theories.

Labert   (2001-11-08 11:03) [11]

Remember the movie "Cards, Money, Two Trunks"? There is an episode where the main characters think how they get 500 pieces of pounds in a week. Well, they also invent various crazy ideas about this ...

PVOzerski   (2001-11-08 11:26) [12]

In my opinion, Newcomer said that he only needs to process the 1 font. It would be nice if both 1 size and 1 style.
Then everything is simple: print all the characters on bitmap "ah, and then stupidly compare by overlay.

Val   (2001-11-08 11:40) [13]

once did something like this - look for information on modeling neural networks and perceptrons, there are several methods for pattern recognition using various techniques, including perceptron self-training

fliz   (2001-11-08 13:13) [14]

but let me doubt that this is a lab.
I once made a living by writing
lab, coursework, etc. And in my opinion this is a graduate
(or course - that is, on __ course__ work on it).

If not, then I would like to know what is it
for such an institution?

Well, on the question itself - my friend, there’s only to encode
and cherish for a week, if not in a hurry. In the meantime, all the algorithms
you will find, you will understand another week will pass. And best of all you
simplify the task - size, rotation angle, etc.
And then just compare the images.

Val   (2001-11-08 13:45) [15]

> fliz © (08.11.01 13: 13)
it also depends on the teacher — in my case, it was a bachelor's degree project (something beyond 5-6 recognition methods), but at the end of the 5 course — we received exactly the same task in the laboratory — do everything in a week :( -real, laboratory it was never handed over by anyone, the teacher, having probably sensed an excess, made a laboratory for those who wanted to increase their grades ...

zero   (2001-11-08 15:38) [16]

> fliz © (08.11.01 13: 13)
We had such a lab designed for a semester, only more complicated.
It was necessary to recognize any text (except handwritten, of course) with the possibility of learning (otherwise you can’t recognize any text).

fliz   (2001-11-08 16:14) [17]

something I feel ashamed of my labors during my student years ... :)

2 zero (08.11.01 15: 38).
interesting to know the results of the work. the phrase "with the possibility of learning"
somehow incomprehensibly tempting. learning what?

Val   (2001-11-08 17:24) [18]

recognition training - in such systems there is feedback

zero   (2001-11-08 17:38) [19]

Look at the beginning of the discussion answer (Natalya © (07.11.01 10: 24))
Я делал прогу примерно по такому алгоритму.
Если встретилась неизвестная буква или иное написание буквы,
то и последовательности чисел будут неизвестные.
В этом случае спрашиваешь что это за буква у пользователя, и связываешь букву с этой последовательностью.
Вначале прогу нужно постоянно обучать, а потом она процентов 90 текста распознает.
Если че еще не ясно спрашивайте. :-)

Новенький   (2001-11-09 00:06) [20]

Спасибо всем кто откликнулся...
Отвечаю сразу на все высказывания:
Это лабораторная работа по Искуственному Интеллекту.Я учусь на 5 курсе Института Информационных Технологий и Коммуникаций Астраханского Государственного Технического Института.
И собираюсьлабу делать(собственно уже пытаюсь делать) на Delphi.Поэтому и написал именно на этот форум
Еще раз спасибо всем и простите, если я нарушил какие-то правила....

модератор   (2001-11-09 09:57) [21]

>Новенький (09.11.01 00:06)
Ничего не нарушил.
Смотри, какую оживленную дискуссию вызвал:-)

zero   (2001-11-09 10:43) [22]

Если лаба по искуственному интелекту, то скорее всего нужно с помощью нейронных сетей, это ближе к теме.
Но исходники свои я тебе уже выслал, лови. Удачи.

Pages: 1 whole branch

Forum: "Grab";
Current archive: 2002.01.08;

Memory: 0.62 MB
Time: 0.032 c