How does offline OMR bubble sheet scanning grade exams in your browser?
Drop a multi-page copier PDF or smartphone photos into the workspace, enter your answer key, and let a local Web Worker score every sheet. Four corner fiducials align tilted pages, jsQR reads the grid layout, and luminance sampling grades marks with 0 bytes uploaded.
Once a scan loads, connected-component thresholding searches the outer 28% border zones for the four solid black fiducial squares printed on printable OMR multiple-choice answer cards. Those four corner coordinates drive a planar projective homography image transformation onto a normalized 1000 × 1400 px grid.
Within that rectified frame, the scanner decodes the top-right F1 QR header automatically. It configures question counts (5 to 200), choice letters (A–D, A–E, or T–F), Student ID columns, and 1-up, 2-up, or 4-up card slots without manual menu setup.
| Export option & view | File extension | Included sheets & tables | Best workflow fit |
|---|---|---|---|
Live browser dashboard |
In-tab view | Student scores + Item analysis + KR-20 |
Immediate post-quiz review and key verification |
Excel workbook |
.xlsx |
Two tabs: Student Scores and Item Analysis |
Department records and spreadsheet filtering |
LMS-ready gradebook CSV |
.csv |
UTF-8 roster with IDs, scores, flags, Q1..Qn |
Direct import or joining into Canvas and Moodle |
Annotated review report |
.pdf |
Cohort stats plus per-student [OK]/[X] cards |
Returning paper score breakdowns to students |
Complete OMR bundle |
.zip |
Combined .xlsx, .csv, and annotated .pdf |
Archiving full exam documentation in one download |
Scoring multi-version Form A through D keys and flagging erasures
Scrambled seating charts only deter copying when grading four separate answer keys stays practical. This bubble sheet scanner reads the shaded Form bubble on every card and routes each student to the matching key automatically.
Entering flexible answer keys and re-scoring mis-keyed items
In single-key mode or after clicking Use multi-version keys (A–D), you can type numbered pairs (1:B, 2:D, 3:A), whitespace-separated letters (B D A C), or a compact string (BDAC). If a student forgets to shade a form bubble, scoring falls back cleanly to Form A / Default answer key.
Discovering a flawed prompt mid-grading never requires re-scanning paper stacks. Edit the key string and click Re-score loaded scans with updated key to refresh every percentage, flag, and distractor count in place before you merge OMR scores into LMS gradebook spreadsheets by Student ID.
How inner-to-halo luminance sampling catches double marks
Phone camera shadows and copier toner gradients ruin naive fixed-threshold pixel counters. Instead, each bubble compares the mean brightness of its inner core (75% radius) against its local paper halo (132% to 195% radius) at a 0.27 contrast threshold normalized against unshaded choices in the same row.
Tip: Check the Flags column inside Student scores right after loading a batch so you can inspect original paper sheets for faint pencil erasures (
A+D) or skipped rows (BLANK) before exporting final grades.
Inside the Annotated review & item report (.pdf), every student receives a dedicated breakdown page arranged across four compact columns. Right answers carry a green [OK] tag, wrong choices show a red [X] beside the expected key letter, and blank or multi-marked rows display an amber [!].
Interpreting KR-20 test reliability and 27% item discrimination
Raw percentages tell you how students performed, whereas psychometrics reveal whether the exam itself measured understanding fairly. Switching from Student scores to Item analysis exposes question-by-question diagnostic stats without external R or spreadsheet formulas.
Evaluating overall exam consistency with KR-20 benchmarks
When at least 3 students complete 2 or more items with non-zero score variance, the top summary card computes the Kuder-Richardson Formula 20 reliability coefficient (KR-20). Classroom unit tests typically target 0.70 to 0.85, while high-stakes department finals aim above 0.80.
A KR-20 value below 0.50 usually signals a short exit ticket covering several unrelated topics, overly tricky wording, or a mis-keyed answer row. Fixing a single mis-entered key letter often raises KR-20 immediately upon re-scoring.
Spotting flawed distractors in the Item analysis table
For batches of 4 or more students, the Discrimination column ranks test-takers by total score and subtracts the correct-response rate of the bottom 27% from the top 27%. Any item scoring below 0.10 triggers a yellow Review badge so you can check for ambiguous phrasing or two defensible choices.
Alongside difficulty and discrimination, the Distractors column tallies exact selection counts (A:2 B:19 C:4 D:1) across every option. On hybrid exams combining machine-scored multiple-choice sections with written free-response essays, instructors often pair OMR item stats with weighted essay and lab rubrics ready for grading to keep scoring transparent.
Scanning multi-page PDFs at 150 to 300 DPI while keeping rosters local
Fast, accurate batch grading depends on clean optical input and strict handling of student identifiers. Both school hallway copiers and mobile cameras work reliably when a few simple capture habits are followed.
Copier document feeder PDFs versus smartphone desk photos
For classes of 20 to 150 learners, feed the entire stack through your office copier’s automatic document feeder at 150 to 300 DPI and drop the resulting multi-page .pdf into step 1. Pages render locally via pdf.js at 145 DPI so a 30-page quiz grades in a couple of seconds.
When photographing sheets on a desk with a phone, keep all four black corner squares inside the camera frame, avoid harsh overhead flash glare, and flatten curled corners. Reviewing classical psychometric item analysis metrics right at your desk lets you address common misconceptions before class ends.
Protecting Student ID numbers with zero cloud uploads
Completed answer sheets link 6-to-10 digit Student IDs directly to academic performance records. Running pdf.js rasterization, homography warping, and spreadsheet generation inside an isolated browser Web Worker ensures zero scan images or Student IDs ever touch an external server.
If you plan to share exported .csv or .xlsx item-response matrices with a curriculum committee, research partner, or external statistician, run the spreadsheet through the FERPA student roster de-identification utility first. Replacing SIS numbers with sequential aliases preserves every item statistic while stripping personally identifiable records.