This paper presents a study on classifying receipts and invoices using Machine Learning. Furthermore, Naïve Bayes Algorithm and the advantages of using it will be discussed.  With information gathered from theory and previous research, I will show how to classify images into a receipt or an invoice. Also, it includes pre-processing images using a variety of pre-processing methods and text extraction using Optical Character Recognition (OCR). Moreover, the necessity of pre-processing images to reach a higher accuracy will be discussed. A result shows a comparison between Tesseract OCR engine and FineReader OCR engine. After embracing much knowledge from theory and discussion, the results showed that combining FineReader OCR engine and Machine Learning is increasing the accuracy of the image classification.

Om två parter med delad ekonomi vill se över och räkna på sina inköp gjort från matvarubutiker finns två alternativ. Att spara alla fysiska kvitton för att sedan manuellt hantera uträkningen eller att använda digitala kvitton, vilket långt ifrån alla matvarubutiker erbjuder. Det finns heller inga bestämmelser mellan företagen kring vart dessa kvitton ska lagras vilket medför att en användare kan behöva logga in på flera olika platser. Användaren har alltså valet att manuellt hantera fysiska kvitton eller att manuellt hantera digitala kvitton, alternativt en blandning av båda. Oavsett kvittots form måste alla kvitton gås igenom för att se vilken person som gjort köpet, vilka eventuella varor som ska plockas bort och vad totalsumman är. Syftet med detta projekt har därför varit att skapa en applikation i Android som med hjälp av OCR-biblioteket Google ML Kit tillåter två användare att hantera sina kvitton. Rapporten undersöker de svårigheter som finns vid textigenkänning samt presenterar de tekniker och metoder som har använts under skapandet av applikationen. Applikationen utvärderades sedan genom att extrahera text från flera olika kvitton. Google’s OCR-bibliotek jämfördes också med Tesseract OCR för att undersöka om valet av ett annat OCR-bibliotek hade kunnat förbättra pålitligheten i kvittoskanningen. Slutresultatet är att applikationen fungerar väl vid korrekt inskanning, det finns dock stora svårigheter att extrahera text från kvitton som avviker från de kvittomallar som använts under implementationen. / For two parties with shared finances who wants to review and count their purchases made from grocery stores, there are two options. To save all physical receipts and then handle the calculation manually or to use digital receipts, which far from all grocery stores offer. There are also no provision between companies on where these receipts should be stored, which means that a user may have to log in at several different locations. The user thus has the choice of manually managing physical receipts or manually managing digital receipts, or in worst case a mixture of both. Regardless of the form of the receipt, all receipts must be reviewed to see which person made the purchase, which items, if any, should be removed and what the total cost is. The aim of this project has therefore been to create an application in Android using the Google ML Kit OCR library that allows two users to manage their receipts. The report examines the difficulties encountered in text recognition and presents the techniques and methods used during the creation of the application. The application was then evaluated by extracting text from several different receipts. Google’s OCR library was also compared with Tesseract OCR to investigate whether the choice of a different OCR library could have improved the reliability of receipt recognition. The final result is a application that works well when a receipt is scanned correctly, however there are significant difficulties in extracting text from receipts that differ from the receipt templates used during implementation.

I dagens flygplan finns det mycket information som på ett snabbt och pålitligt sätt behöver förmedlas till piloten via instrument på flera skärmar i cockpit. Att verifiera att skärmarna visar korrekt data för olika indata är ett tidskrävande och monotont arbete. Därför undersöker Saab möjligheten att automatisera delar av arbetet. Examensarbetet undersöker genom praktiskt implementation ifall det är möjligt att automatisera bildanalysen med hjälp av programmen OpenCV och Tesseract. Resultatet visade att det går att enkelt konstruera tester för att automatiskt identifiera oönskade förändringar i den implementerade instrumentingen.

-&gt;Context. Optical Character Recognition (OCR), having a computer recognize text from an image, is not as intuitive as human recognition. Even small (to human eyes) degradations can thwart the OCR result. The problem is that random unknown degradations are unavoidable in a real-world setting. -&gt;Objectives. The noise tolerance of Tesseract, a state-of-the-art OCR engine, is evaluated in relation to how well it handles salt and pepper noise, a type of image degradation. Noise tolerance is measured as the percentage of aberrant pixels when comparing two images (one with noise and the other without noise). -&gt;Methods. A novel systematic approach for finding the noise tolerance of an OCR engine is presented. A simulated environment is developed, where the test parameters, called test cases (font, font size, text string), can be modified. The simulation program creates a text string image (white background, black text), degrades it iteratively using salt and pepper noise, and lets Tesseract perform OCR on it, in each iteration. The iteration process is stopped when the comparison between the image text string and the OCR result of Tesseract mismatches. -&gt;Results. Simulation results are given as changed pixels percentage (noise tolerance) between the clean text string image and the text string image the degradation iteration before Tesseract OCR failed to recognize all characters in the text string image. The results include 14400 test cases: 4 fonts (Arial, Calibri, Courier and Georgia), 100 font sizes (1-100) and 36 different strings (4*100*36=14400), resulting in about 1.8 million OCR attempts performed by Tesseract. -&gt;Conclusions. The noise tolerance depended on the test parameters. Font sizes smaller than 7 were not recognized at all, even without noise applied. The font size interval 13-22 was the peak performance interval, i.e. the font size interval that had the highest noise tolerance, except for the only monospaced font tested, Courier, which had lower noise tolerance in the peak performance interval. The noise tolerance trend for the font size interval 22-100 was that the noise tolerance decreased for larger font sizes. The noise tolerance of Tesseract as a whole, given the experiment results, was circa 6.21 %, i.e. if 6.21 % of the pixel in the image has changed Tesseract can still recognize all text in the image. / <p>42</p>

<p>Denna rapport handlar om att ta fram ett OCR-verktyg för digitalisering av pappersdokument. Krav på detta verktyg är att bland annat det ska vara kompatibelt med Linux, det ska kunna ta kommandon via kommandoprompt och dessutom ska det kunna hantera skandinaviska tecken.</p><p>Tolv OCR-verktyg granskades, sedan valdes tre verktyg ut; Ocrad, Tesseract och OCR Shop XTR. För att testa dessa scannades två dokument in och digitaliserades i varje verktyg.</p><p>Resultatet av testerna är att Tesseract är de verktyget som är mest precist och Ocrad är det verktyget som är snabbast. OCR Shop XTR visar på sämst resultat både i tidtagning och i antal korrekta ord.</p> / <p>This report is about finding OCR software for digitizing paper documents. Requirements were to include those which were compatible with Linux, being able to run commands via the command line and also being able to handle the Scandinavian characters.</p><p>Twelve OCR softwares were reviewed, and three softwares were chosen; Ocrad, Tesseract and OCR Shop XTR. To test these, two document were scanned and digitized in each tool.</p><p>The results of the tests are that Tesseract is the tool which is the most precise and Ocrad is the tool which is the fastest. OCR Shop XTR shows the worst results both in timing and number of correct words.</p>

Denna rapport handlar om att ta fram ett OCR-verktyg för digitalisering av pappersdokument. Krav på detta verktyg är att bland annat det ska vara kompatibelt med Linux, det ska kunna ta kommandon via kommandoprompt och dessutom ska det kunna hantera skandinaviska tecken. Tolv OCR-verktyg granskades, sedan valdes tre verktyg ut; Ocrad, Tesseract och OCR Shop XTR. För att testa dessa scannades två dokument in och digitaliserades i varje verktyg. Resultatet av testerna är att Tesseract är de verktyget som är mest precist och Ocrad är det verktyget som är snabbast. OCR Shop XTR visar på sämst resultat både i tidtagning och i antal korrekta ord. / This report is about finding OCR software for digitizing paper documents. Requirements were to include those which were compatible with Linux, being able to run commands via the command line and also being able to handle the Scandinavian characters. Twelve OCR softwares were reviewed, and three softwares were chosen; Ocrad, Tesseract and OCR Shop XTR. To test these, two document were scanned and digitized in each tool. The results of the tests are that Tesseract is the tool which is the most precise and Ocrad is the tool which is the fastest. OCR Shop XTR shows the worst results both in timing and number of correct words.

Companies often process invoices manually, therefore automation could reduce manual labor. The aim of this thesis is to evaluate which OCR-engine, Tesseract or OCRopus, performs best at interpreting invoices. This thesis also evaluates if it is possible to use machine learning to automatically process invoices based on previously stored data. By interpreting invoices with the OCR-engines, it results in the output text having few spelling errors. However, the invoice structure is lost, making it impossible to interpret the corresponding fields. If Naïve Bayes is chosen as the algorithm for machine learning, the prototype can correctly classify recurring invoice lines after a set of data has been processed. The conclusion is, neither of the two OCR-engines can interpret the invoices to plain text making it understandable. Machine learning with Naïve Bayes works on invoices if there is enough previously processed data. The findings in this thesis concludes that machine learning and OCR can be utilized to automatize manual labor. / Företag behandlar oftast fakturor manuellt och en automatisering skulle kunna minska fysiskt arbete. Målet med examensarbetet var att undersöka vilken av OCR-läsarna, Tesseract och OCRopus som fungerar bäst på att tolka en inskannad faktura. Även undersöka om det är möjligt med maskininlärning att automatiskt behandla fakturor utifrån tidigare sparad data. Genom att tolka text med hjälp av OCR-läsarna visade resultaten att den producerade texten blev språkligt korrekt, men att strukturen i fakturan inte behölls vilket gjorde det svårt att tolka vilka fält som hör ihop. Naïve Bayes valdes som algoritm till maskininlärningen och resultatet blev en prototyp som korrekt kunde klassificera återkommande fakturarader, efter att en mängd träningsdata var behandlad. Slutsatsen är att ingen av OCR-läsarna kunde tolka fakturor så att resultatet kunde användas vidare, och att maskininlärning med Naïve Bayes fungerar på fakturor om tillräckligt med tidigare behandlad data finns. Utfallet av examensarbetet är att maskininlärning och OCR kan användas för att automatisera fysiskt arbete.

This paper tackles the challenge of how to compare floor plans with each other. A lot of different methods were used to analyze floor plan images, such as different kinds of pixel-based breadth-first search algorithms for finding walls, doors, and windows. Python-tesseract was used to read text labels in the floor plan, which was of great use when rooms were to be identified. The extracted information from over 1000 floor plans was then used to create a comparison program, which spits out the most similar floor plans to any given floor plan. The results of the extraction part were pretty good for most of the floor plans. Walls, doors, and windows were often accurately found, and the room identification worked very well compared to other known methods. Using the extracted data to find similar floor plans worked splendidly. The extraction part of the project had its flaws and can be improved, but even so, this method of assessing similarity between floor plans works very well. / Den här rapporten tacklar problemet och försöker ge ett svar på hur man kan jämföra planritningar med varandra. Flera olika metoder användes för att analysera planritningar, exempelvis olika typer av pixelbaserade sökalgoritmer för att hitta planritningens väggar, dörrar och fönster. Python-tesseract användes också för att läsa textetiketter i planritningen, vilket var till stor nytta när rummen skulle identifieras. Den extraherade informationen från över 1 000 planritningar användes sedan för att skapa ett jämförelseprogram, som spottar ut de 10 mest liknande planritningarna till en given planritning. Resultatet av extraheringsdelen var väldigt bra för de flesta planritningarna. Se exempelbilden nedan. Väggar, dörrar och fönster hittades ofta korrekt och rumidentifieringen fungerade mycket bra jämfört med andra kända metoder. Att använda den extraherade information för att sedan hitta liknande planlösningar fungerade utmärkt. Extraheringsdelen av projektet hade sina brister och kan förbättras, men trots det fungerar denna metod för att jämföra planlösningar väldigt bra.

Today’s airplanes and modern cars are equipped with displays to communicate important information to the pilot or driver. These displays needs to be tested for safety reasons; displays that fail can be a huge safety risk and lead to catastrophic events. Today displays are tested by checking the output signals or with the help of a person who validates the physical display manually. However this technique is very inefficient and can lead to important errors being unnoticed. MindRoad AB is searching for a solution where validation of the display is made from a camera pointed at it, text and numbers will then be recognized using a computer vision algorithm and validated in a time efficient and accurate way. This thesis compares the three different text detection algorithms, EAST, SWT and Tesseract to determine the most suitable for continued work. The chosen algorithm is then optimized and the possibility to develop a program which meets MindRoad ABs expectations is investigated. As a result several algorithms were combined to a fully working program to detect and recognize text in industrial displays.

Deep learning has become ubiquitous as part of Optical Character Recognition (OCR), but there are few examples of research into whether the two technologies are feasible for deployment on a mobile platform. This study examines which particular method of OCR would be best suited for a mobile platform in the specific context of a prescription medication label scanner. A case study using three different methods of OCR – classic computer vision techniques, standard deep learning and specialised deep learning – tested against 100 prescription medicine label images shows that the method that provides the best combination of accuracy, speed and resource using has proven to be standard seep learning, or Tesseract 4.1.1 in this particular case. Tesseract 4.1.1 tested with 76% accuracy with a further 10% of results being one character away from being accurate. Additionally, 9% of images were processed in less than one second and 41% were processed in less than 10 seconds. Tesseract 4.1.1 also had very reasonable resource costs, comparable to methods that did not utilise deep learning.