Character Recognition
显示 更早的评论
First, I need to count the Pixels in an image(Ex:-image of Letter A) then I need to check whether the image is accurate(check whether "image of Letter A is correct") , can I do it by counting pixels of the image and check whether it is correct...can anyone know it please Reply or post a sample mathlab code for that
采纳的回答
Walter Roberson
2011-4-29
0 个投票
What difference is there between "count the Pixels in an image [...] then I need to check whether the image is accurate" compared to "can I do it by counting pixels of the image and check whether it is correct" ?
If the question is whether you can determine whether a particular character is correct just by counting its pixels, the answer is NO.
6 个评论
prathiba
2011-4-29
Walter Roberson
2011-4-29
One thing you have not indicated about this project is whether your hardware is pressure sensitive or not. Recognition of strokes over time is, I gather, easier than recognition from pixels alone.
There are many different ways to recognize characters; the field is known as Optical Character Recognition, or OCR, and people have been publishing papers on it for a minimum of 40 years.
On the other hand, typical OCR involves finding the best character match for what is presented, rather than in deciding whether a character has been drawn "correctly" or not. Whether a character has been drawn "correctly" depends entirely upon the rules that have been set up for correctness. For example, for a capital-A, people can read it if the horizontal bar is too long or too short on one or both sides; even the angled strokes together with just a dot in the middle would generally be readable by people. As, however, you are trying to train the children to certain standards, you have to decide not "what could be recognized" but instead "what is allowable for teaching purposes". What is too shallow, what is too steep, what slant is permitted, how consistent does the baseline have to be, and so on. Typeset capital-A often has small horizontal bars at the very peak of the A and at the bottom of each of the legs: are the children _required_ to put those in, or are they required to leave them out, or is putting one on top unacceptable but on the legs is okay, or is it the case that it is acceptable to have them on neither leg or on both legs but not acceptable to put the bars on just one of the legs?
You (or someone) needs to decide what standards are to be set for teaching purposes, right down to details like the acceptable length (or proportion) of optional horizontal bars on the legs of the "A". The question is not of robust OCR under a variety of handwriting or typefaces, but instead is a matter of what will be allowed.
prathiba
2011-4-30
Walter Roberson
2011-4-30
Comparing images in the manner you are hinting at would very likely fail. For example, suppose the child drew the letter _exactly_ as taught, but drew it slightly smaller than your reference image: you would not match. Now suppose that the child drew the image exactly the same size (to the pixel) as the reference image, but their hand wavered a little and one of the strokes was not _exactly_ straight. Now suppose the child used the curved form of "A" as is often seen in books.
Hardware tablets are, in a way, like mice, in that neither internally stores an "image" of which pixels are passed over: both tablets and mice instead report current position information and push or click information. Some hardware tablets can also record pressure. For both tablets and mice, to draw a character, _something_ needs to interpret the the position and up/down information and draw that in to a matrix that can then be interpreted as an image of a character. There are MATLAB routines that can assemble a series of line segments from mice, which should be relatively easily convertible for use with tablets. There are then a few different ways of drawing the line segments in to an image matrix.
Getting the input is the easy part. Interpreting the input is the hard part. No-one is going to be able to give you good suggestions about how to interpret the input as "correct" or not until you define the parameters of what will be *accepted* as "correct" -- a quite different task than recognizing the closest match.
If you do have input right from the tablet rather than just the final character images, then the following might be of interest to you: http://hwr.nici.kun.nl/unipen/
prathiba
2011-4-30
Walter Roberson
2011-4-30
Sorry, I have given up on the strategy of listing all *possible* answers to a question and letting the other person figure out which one is the most suitable answer.
Research character recognition in Google Scholar. Any technique you can find there can be implemented in MATLAB. Some techniques are easier to implement in MATLAB than others are.
You need to define what it means for a character to be written "correctly" *first*, and _then_ you figure out what technique can be used to decide whether any particular character is correct or not.
I am not going to let you off on this one. If *you* cannot describe what you look at to determine whether a character is written "correctly" for teaching purposes, then *we* cannot guide you to suitable technologies to distinguish between "correct" and "incorrect".
更多回答（0 个）
类别
在 帮助中心 和 File Exchange 中查找有关 Language Support 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!
Translated by 
