import ocr ares

src-tauri/src/ocr.rs

@@ -1,4 +1,4 @@
-use image::RgbaImage;
+use image::{GenericImageView, Rgba, RgbaImage};
 use ort::session::{Session, builder::GraphOptimizationLevel};
 use ort::value::Value;
 use std::path::Path;
@@ -31,7 +31,26 @@ pub fn run_ocr_detection(
     // And usually resized to multiple of 32. Limit max size for speed.
     let max_side = 1600; // Increase resolution limit
     let (orig_w, orig_h) = input_image.dimensions();
+
+    // --- CROP to top 5% and bottom 5% ---
+    let crop_height = (orig_h as f64 * 0.05).ceil() as u32;
+    let mut masked_image = RgbaImage::from_pixel(orig_w, orig_h, Rgba([0, 0, 0, 255]));
     
+    if crop_height > 0 {
+        // Copy top part
+        let top_view = input_image.view(0, 0, orig_w, crop_height).to_image();
+        image::imageops::replace(&mut masked_image, &top_view, 0, 0);
+        
+        // Copy bottom part
+        let bottom_y = orig_h.saturating_sub(crop_height);
+        let bottom_view = input_image.view(0, bottom_y, orig_w, crop_height).to_image();
+        image::imageops::replace(&mut masked_image, &bottom_view, 0, bottom_y as i64);
+    } else {
+        // If image is very short, just use the original
+        masked_image = input_image.clone();
+    }
+    // --- End CROP ---
+
     let mut resize_w = orig_w;
     let mut resize_h = orig_h;
     
@@ -50,7 +69,7 @@ pub fn run_ocr_detection(
     resize_w = resize_w.max(32);
     resize_h = resize_h.max(32);
 
-    let resized = image::imageops::resize(input_image, resize_w, resize_h, image::imageops::FilterType::Triangle);
+    let resized = image::imageops::resize(&masked_image, resize_w, resize_h, image::imageops::FilterType::Triangle);
     
     let channel_stride = (resize_w * resize_h) as usize;
     let mut input_data = Vec::with_capacity(1 * 3 * channel_stride);