Fix Hardsubx OCR (#1741)

* fix: hardsubx segmentation fault

* fix: hardsubx garbage output

* chore: enable hardsubx on test builds

Author: hrideshmg

.github/workflows/build_linux.yml

@@ -26,10 +26,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Install dependencies
-      run: sudo apt update && sudo apt-get install libgpac-dev libtesseract-dev
+      run: sudo apt update && sudo apt-get install libgpac-dev libtesseract-dev libavcodec-dev libavdevice-dev libx11-dev libxcb1-dev libxcb-shm0-dev
     - uses: actions/checkout@v4
     - name: build
-      run: ./build
+      run: ./build -hardsubx
       working-directory: ./linux
     - name: Display version information
       run: ./ccextractor --version

docs/CHANGES.TXT

@@ -1,5 +1,6 @@
 1.0 (to be released)
 -----------------
+- Fix: HardSubX OCR on Rust
 - Removed the Share Module
 - Fix: Regression failures on DVD files
 - Fix: Segmentation faults on MP4 files with CEA-708 captions

src/rust/src/hardsubx/decoder.rs

@@ -40,22 +40,25 @@ pub unsafe fn dispatch_classifier_functions(ctx: *mut lib_hardsubx_ctx, im: *mut
     // function that calls the classifier functions
     match (*ctx).ocr_mode {
         0 => {
-            let ret_char_arr = get_ocr_text_wordwise_threshold(ctx, im, (*ctx).conf_thresh);
-            ffi::CStr::from_ptr(ret_char_arr)
-                .to_string_lossy()
-                .into_owned()
-        }
-        1 => {
-            let ret_char_arr = get_ocr_text_letterwise_threshold(ctx, im, (*ctx).conf_thresh);
+            let ret_char_arr = get_ocr_text_simple_threshold(ctx, im, (*ctx).conf_thresh);
             let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
             match text_out_result {
                 Ok(T) => T,
                 Err(_E) => "".to_string(),
             }
         }
-
+        1 => {
+            let ret_char_arr = get_ocr_text_wordwise_threshold(ctx, im, (*ctx).conf_thresh);
+            if ret_char_arr.is_null() {
+                "".to_string()
+            } else {
+                ffi::CStr::from_ptr(ret_char_arr)
+                    .to_string_lossy()
+                    .into_owned()
+            }
+        }
         2 => {
-            let ret_char_arr = get_ocr_text_simple_threshold(ctx, im, (*ctx).conf_thresh);
+            let ret_char_arr = get_ocr_text_letterwise_threshold(ctx, im, (*ctx).conf_thresh);
             let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
             match text_out_result {
                 Ok(T) => T,
@@ -66,7 +69,6 @@ pub unsafe fn dispatch_classifier_functions(ctx: *mut lib_hardsubx_ctx, im: *mut
         _ => {
             eprintln!("Invalid OCR Mode");
             exit(EXIT_MALFORMED_PARAMETER);
-            // "".to_string()
         }
     }
 }
@@ -113,7 +115,7 @@ pub unsafe extern "C" fn _process_frame_white_basic(
     let mut gray_im: *mut Pix = pixConvertRGBToGray(im, 0.0, 0.0, 0.0);
     let mut sobel_edge_im: *mut Pix =
         pixSobelEdgeFilter(gray_im, L_VERTICAL_EDGES.try_into().unwrap());
-    let mut dilate_gray_im: *mut Pix = pixDilateGray(sobel_edge_im, 21, 1);
+    let mut dilate_gray_im: *mut Pix = pixDilateGray(sobel_edge_im, 21, 11);
     let mut edge_im: *mut Pix = pixThresholdToBinary(dilate_gray_im, 50);
 
     let mut feat_im: *mut Pix = pixCreate(width, height, 32);

src/rust/src/hardsubx/imgops.rs

@@ -13,10 +13,11 @@ pub extern "C" fn rgb_to_hsv(R: f32, G: f32, B: f32, H: &mut f32, S: &mut f32, V
 
 #[no_mangle]
 pub extern "C" fn rgb_to_lab(R: f32, G: f32, B: f32, L: &mut f32, a: &mut f32, b: &mut f32) {
-    let rgb = Srgb::new(R, G, B);
+    // Normalize input RGB from 0-255 to 0.0-1.0
+    let rgb = Srgb::new(R / 255.0, G / 255.0, B / 255.0);
 
-    // This declaration sets the white-point as per the D65 standard
-    let lab_rep = Lab::<palette::white_point::D65, f32>::from_color(rgb);
+    // Convert from sRGB to Lab (D65 white point is default)
+    let lab_rep = Lab::from_color(rgb);
 
     *L = lab_rep.l;
     *a = lab_rep.a;