ocr.rar_OCR_文字识别

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% %% %% Prof. Sclaroff's CS585 Image avd Video Processing %% %% Project ONE %% %% C H A R A C T E R R E C O G N I T I O N %% %% %% %% by Stanislav Rost %% %% ID: 31764117 %% %% %% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function result = vowel(filename) % VOWEL.M % % function result = vowel(filename) % % Arguments: name of the file containing scanned letters % % Vowel.m performs vowel recognition using moments, part ratios and % compactness. % % As a result, the function shows an image with letters O marked % blue, E magenta, A red, U green. All other letters are white/gray. % % Assumptions: % - if the paragraph is tilted, then it is rectangular % - no more than two letters are joined at any instance of joined letters imageData = im2bw(imread(filename,'tif'),0.5); imageData = ~imageData; % Find out average height of letters avgLabel = bwlabel(imageData, 4); avgLetters = imfeature(avgLabel, 'BoundingBox'); numAvg = max(avgLabel(:)); avgHeight = 0; for i = 1:numAvg, avgHeight = avgHeight + avgLetters(i).BoundingBox(4); end avgHeight = round(avgHeight / numAvg); clear avgLetters % #################### ROTATOR ############################ % Rotate the image so that letters are properly oriented % This is how it's done: we use the find out imfeature's Orientation % for the entire paragraph (so the assumption here is that the paragraph % is wider than it is tall), and then rotate by that value... % only if the image is tilted, which can be checked by breaking the image in 2, % and then seeing if the top boundaries on their bounding boxes are about the % same height origFeature = imfeature(double(imageData), 'BoundingBox', 'Orientation'); angle = - origFeature.Orientation; midpoint = floor(origFeature.BoundingBox(3)/2); leftHalf = imageData(:,1:midpoint); rightHalf = imageData(:,(midpoint+1):end); leftFeature = imfeature(double(leftHalf), 'BoundingBox'); rightFeature = imfeature(double(rightHalf), 'BoundingBox'); % If the distance between upper boundaries of bounding boxes % is bigger than half the average letter size, rotate image. if (abs(leftFeature.BoundingBox(2) - rightFeature.BoundingBox(2)) > avgHeight/2 ) disp('Rotation accepted'); imageData = imrotate(imageData, angle, 'nearest'); end % ################### LABEL 1 ############################## imageData = bwlabel(imageData, 4); % #################### SPLITTER ############################ % Split the letters that need to be split % First, calculate the averge width of letters splitLetters = imfeature(imageData, 'BoundingBox', 'Image'); numSplit = max(imageData(:)); avgWidth = 0; for i = 1:numSplit, avgWidth = avgWidth + splitLetters(i).BoundingBox(3); end avgWidth = round(avgWidth / numSplit); % Now, go through all the letters and if their width % is bigger than 1.5 average then split them. for i = 1:numSplit, if splitLetters(i).BoundingBox(3) > 1.5 * avgWidth [ rows columns ] = size(splitLetters(i).Image); % Draw vertical line splitLetters(i).Image(:, avgWidth ) = 0; splitLetters(i).Image(:, splitLetters(i).BoundingBox(3) - avgWidth) = 0; if (columns>avgWidth*2.5) splitLetters(i).Image(:, columns - avgWidth +2 ) = 0; end for k = 1:2, splitLetters(i).Image =erode(splitLetters(i).Image, ... [ 0,1,0;0,1,0 ]); splitLetters(i).Image = bwmorph(splitLetters(i).Image, 'spur'); end % insert the split image back into the original image xoffset = round(splitLetters(i).BoundingBox(1)); yoffset = round(splitLetters(i).BoundingBox(2)); % Put the image back into training data for c = 1:splitLetters(i).BoundingBox(3), for d = 1:splitLetters(i).BoundingBox(4), imageData(yoffset+d-1, xoffset+c-1) = ... splitLetters(i).Image(d,c); end end end end imageData = im2bw(imageData, 0.5); imageData = bwlabel(imageData, 4); % #################### RECOGNITION LOOP ################ % Relabel everything because it might be rotated imageLetters = imfeature(imageData,'Image', 'Area', 'EulerNumber', 'BoundingBox'); imageNumLetters = max(imageData(:)); % Load recognizable letter data from file(s) letterName(1) = 'o'; letterName(2) = 'e'; letterName(3) = 'a'; letterName(4) = 'u'; load 'o.mat' meanO covO letterObj(1) = struct('Name', 'o', ... 'Mean', meanO, ... 'Covariance', covO); load 'e.mat' meanE covE letterObj(2) = struct('Name', 'e', ... 'Mean', meanE, ... 'Covariance', covE); load 'a.mat' meanA covA letterObj(3) = struct('Name', 'a', ... 'Mean', meanA, ... 'Covariance', covA); load 'u.mat' meanU covU letterObj(4) = struct('Name', 'u', ... 'Mean', meanU, ... 'Covariance', covU); % Save a copy of a colormap cmap = colormap; % A spinning tick mark indicates that processing is underway tickArray = [ '-' '/' '|' '\' '-' '/' '|' '\' ]; % Main recognition loop % newColors will hold the number of colors that was added as a result of splitting newColors = 0; % Main recognition loop - go through each letter (or joined sequence of letters) for k = 1:imageNumLetters, % Output the tick mark temp = mod(k,8); if temp == 0 temp = 8; end disp(tickArray(temp)); % It's a single letter % Compare each letter to be recognized to each letter matchLetter = letter_compare(letterObj, imageLetters(k)); % Highlight the letter switch matchLetter case 1, palentry = [ 0.0 0.0 1.0 ]; case 2, palentry = [ 1.0 0.0 1.0 ]; case 3, palentry = [ 1.0 0.0 0.0 ]; case 4, palentry = [ 0.0 1.0 0.0 ]; otherwise, palentry = [ 0 0 0 ]; end cmap(k,:) = palentry; end % End main recognition loop % Shift the data values up to make room for black background imageData = imageData +1; % Add the black color into the palette tempMap = [ 1,1,1; cmap ]; % Show results figure; image(imageData); colormap(tempMap);