Batch Convert 30,000 Images to Excel with Python: Automating 300,000 Data Entries

In a typical office setting, operations and finance staff often face the painful task of aggregating scattered image‑based reports. The author received a request to process more than 30,000 report images containing roughly 300,000 raw data rows and to consolidate all tables into a single Excel worksheet.

Manual entry was infeasible, and the built‑in WPS image‑to‑table feature crashed when attempting batch imports. An alternative workflow—converting images to PDF and then parsing tables—yielded only about 95% extraction accuracy, leaving tens of thousands of rows to be corrected manually.

Key insight: WPS can recognize a single image with 100% precision, but it lacks batch automation capabilities. This limitation became the core reason for choosing Python for automation.

Initially the author tried a low‑code tool (WorkBuddy) to invoke WPS’s OCR API, but the tool could not access the native OCR interface and performed poorly on small batches. Turning to the AI assistant DeepSeek, the author received a concrete plan: develop a Python script that automates image preprocessing, OCR table extraction, and data aggregation.

Starting from zero, the author installed Python 3.12 following AI‑generated instructions, copied the initial code, and successfully produced structured table data on the first run. However, the first script’s recognition accuracy fell short of business standards.

Discovering a colleague’s custom image‑table parsing code written for Python 3.8, the author encountered version‑dependency conflicts. By feeding the source code to the AI, they adapted the script to run under the local Python 3.12 environment, fixing import errors and aligning dependencies.

During manual trial‑and‑error, the author noticed that dense header fields at the top of each image often caused misrecognition. To mitigate this, they added a Python batch‑image‑cropping step that trimmed the noisy header region before OCR. The original colleague’s code used a “image partition fixed‑point recognition” approach, dividing each report into multiple zones and limiting OCR to specific field ranges, which greatly reduced column‑shift errors. The author re‑implemented the partition rules for the cropped images, iterating with the AI until the script handled the new image dimensions correctly.

After these optimizations, the final script processed all 30,000 images automatically, achieving a stable table‑recognition accuracy above 99%, fully meeting the business’s precision requirements.

All extracted data were merged into a single Excel file. To further reduce manual verification, the author imported the workbook into WorkBuddy, defined custom data‑validation rules, and let the tool automatically flag anomalies, generate issue lists by row number and error type, and pinpoint dirty data, dramatically cutting review effort.

The experience reinforced the author’s view that AI‑assisted Python automation is most valuable when applied to repetitive, standardized tasks: it eliminates human error and compresses labor time by orders of magnitude, highlighting Python’s indispensable role in workplace data processing.

Future work

The project will be split into four core code modules—batch image cropping, OCR table extraction, Excel data merging, and abnormal data validation—with detailed source‑code walkthroughs planned for upcoming articles.