import { createWorker, Worker } from "tesseract.js";
import { OcrResult, OcrElement } from "../types/ocr.js";
let worker: Worker | null = null;
async function getWorker(): Promise<Worker> {
if (!worker) {
worker = await createWorker("eng");
}
return worker;
}
export async function extractText(imagePath: string): Promise<OcrResult[]> {
const w = await getWorker();
const { data } = await w.recognize(imagePath, {}, { blocks: true });
const results: OcrResult[] = [];
// Navigate through blocks -> paragraphs -> lines -> words
if (data.blocks) {
for (const block of data.blocks) {
for (const paragraph of block.paragraphs) {
for (const line of paragraph.lines) {
for (const word of line.words) {
if (word.text.trim()) {
results.push({
text: word.text,
confidence: word.confidence / 100, // Normalize to 0-1
bounds: {
x0: word.bbox.x0,
y0: word.bbox.y0,
x1: word.bbox.x1,
y1: word.bbox.y1,
},
});
}
}
}
}
}
}
return results;
}
export async function terminateOcr(): Promise<void> {
if (worker) {
await worker.terminate();
worker = null;
}
}
export function searchText(ocrResults: OcrResult[], searchTerm: string): OcrElement[] {
const lowerSearch = searchTerm.toLowerCase();
const matches = ocrResults.filter(
(result) => result.text.toLowerCase().includes(lowerSearch)
);
// Index represents position in filtered matches (0, 1, 2...) for use with elementIndex tap
// This is intentional - users tap by match index, not original OCR result position
return matches.map((match, index) => ({
index,
text: match.text,
bounds: `[${match.bounds.x0},${match.bounds.y0}][${match.bounds.x1},${match.bounds.y1}]`,
center: {
x: Math.round((match.bounds.x0 + match.bounds.x1) / 2),
y: Math.round((match.bounds.y0 + match.bounds.y1) / 2),
},
confidence: match.confidence,
}));
}
