Enhance Python-based OCR Readiness Platform

Project: Fix and Enhance Existing OCR Readiness Assessment Platform Hello, I have an existing OCR Readiness Assessment Platform developed using Python, Streamlit, OpenCV and Tesseract OCR. Some features, especially image refinement and extracted-text accuracy, are not working correctly. I need an experienced Python developer to fix and improve the existing application without breaking its current functionality. I will share the project repository after discussing the project with a suitable freelancer. 1. Project Overview The platform evaluates document image quality before OCR processing. It calculates 10 quality factors, generates an overall OCR Readiness Score, extracts text using Tesseract OCR, and displays recommendations and analysis. The 10 factors are: 1. Noise 2. Resolution 3. Blur 4. Contrast 5. Stroke Width 6. Text Density 7. Matra Continuity 8. Zone Integrity 9. Connected Component Stability (CC Stability) 10. Skew Penalty The application already has image upload, quality-score cards, factor information, recommendations, OCR extraction, confidence highlighting, refinement previews and other dashboard features. I want the existing application fixed and enhanced, not rebuilt from scratch. 2. Image Refinement for All 10 Factors I need a properly working Refine button for all 10 factors wherever refinement is technically appropriate. Refinement eligibility - Score below 20: Disable refinement and recommend uploading a better-quality image or rescanning the document. - Score from 20 to 80, inclusive: Allow refinement. - Score above 80: Show that refinement is not needed. These rules apply to individual factor scores, not the overall OCR Readiness Score. Expected behavior when Refine is clicked 1. Identify the actual image-quality problem associated with the selected factor. 2. Apply a suitable image-processing technique for that specific problem. 3. Generate an actual refined image. 4. Recalculate the target factor and all other factor scores using the existing scoring pipeline. 5. Recalculate the overall OCR Readiness Score. 6. Show the original and refined images side by side. 7. Display before-and-after scores and validation results. The refined image must show a meaningful visual difference related to the selected factor. Simply changing the score, label, recommendation or UI is unacceptable. Examples of refinement targets: - Noise: reduce unwanted noise. - Resolution: apply appropriate enhancement without falsely claiming to recover missing detail. - Blur: address blur using a technically suitable method. - Contrast: improve inadequate contrast. - Stroke Width: correct excessively thin or thick strokes where possible. - Text Density: address density-related defects only when justified. - Matra Continuity: preserve and repair appropriate Devanagari matras and Shirorekha. - Zone Integrity: address actual structural defects in text regions. - CC Stability: address unwanted noise components, fragmentation and irregular connected components. - Skew Penalty: correct actual document skew. The developer must inspect the existing implementations and select appropriate transformations. Do not blindly apply the same operation to every image. 3. Mandatory Safety Checks Accept a refined candidate only if: - The target factor improves by at least 0.5 points. - No non-target factor decreases by more than 5 points. - The overall OCR Readiness Score strictly increases. - The actual image transformation is appropriate and does not unnecessarily damage the text. If these conditions fail, reject the candidate and display a clear message such as "No safe refinement was found for this factor." Do not manipulate scoring formulas or artificially increase scores to pass validation. If no safe improvement is technically possible, report that honestly. 4. Multiple-Factor Refinement Add a feature allowing users to select multiple eligible factors and refine them in one workflow. For example, users may select Blur, Contrast and Stroke Width, or CC Stability and Zone Integrity.

Requirements

- Allow selection of multiple eligible factors. - Apply the appropriate transformation for each selected factor. - Process refinements in a controlled sequence. - Use the currently accepted image as the input for the next refinement. - Recalculate all factor scores and overall readiness after every accepted step. - Apply the same safety checks to every step. - Reject unsafe candidates without replacing the last accepted image. - Display the original image, final accepted image and individual score changes. If one refinement fails, report it clearly and allow other selected factors to be attempted where appropriate. Batch refinement must not bypass safety checks. 5. Fix Tesseract OCR and Extracted Text The Extracted Text section sometimes displays incorrect or garbled text, incomplete words, unexpected characters and misplaced symbols. Please inspect and fix the existing OCR pipeline, including: - Tesseract configuration and language data. - Hindi/Devanagari and English OCR support where appropriate. - Image preprocessing, scaling, thresholding and page segmentation. - Incorrect word segmentation and missing characters. - Word-confidence values and confidence highlighting. - Ensuring OCR processes the correct uploaded, selected or accepted refined image. Extracted text must come from actual Tesseract processing. Do not hardcode or fabricate text. Test the results using suitable Hindi and English document images. 6. Refinement Preview and Controls Provide: - Original/current active image and refined candidate. - Before-and-after target factor scores. - Before-and-after overall OCR Readiness Scores. - A table showing changes in all 10 factors. - Safety-validation results. - Apply, Cancel, Revert and Download functionality. Apply should accept the candidate. Cancel should discard it. Revert should restore the appropriate previous/original image. Download should provide the accepted refined image. 7. Preserve Existing Func