Better Japanese post-processing

.claude/settings.local.json

@@ -10,7 +10,8 @@
       "Bash(dotnet --version)",
       "Bash(dotnet build:*)",
       "Bash(dotnet test:*)",
-      "Bash(gh pr list:*)"
+      "Bash(gh pr list:*)",
+      "Bash(\"/mnt/c/Program Files/dotnet/dotnet.exe\" test Tests/Tests.csproj --filter \"FullyQualifiedName~OcrJapaneseImage\" --logger \"console;verbosity=detailed\")"
     ],
     "deny": []
   }

Tests/OcrTests.cs

@@ -1,4 +1,4 @@
-using System.Drawing;
+using System.Drawing;
 using System.IO;
 using System.Text;
 using System.Text.Json;
@@ -90,6 +90,21 @@ BANK CHARGES	344	670	326	49%
 TOTAL EXPENDITURES	1,016,684	2,065,620	1,048,936	51%
 REVENUES OVERY(UNDER) EXPENDITURES	$9,749	$0	$9,749	N/A";
 
+    private const string jaWordBorders = @".\TextFiles\ja-word-borders.json";
+    private const string jaExpectedResult = """
+        くろ　からだ　しつ
+        黒ごまは体にいいです。タンパク質やカルシウムが
+        かみ　くろ　こうか
+        たくさんあります。髪を黒くする効果もあります。
+        くろ　あぶら　はだ　かみ　りょうり
+        黒ごま油は肌や髪に使います。料理にも使います。
+        かゆ　た　からだ
+        お粥やデザートに入れます。でも、食べすぎると体 た
+        によくないです。少しずつ食べましょう。
+        """;
+
+    private const string jaTestImagePath = @".\Images\ja-黒くろごまのちから.png";
+
     [WpfFact]
     public async Task OcrFontSampleImage()
     {
@@ -362,4 +377,26 @@ public async Task GetTesseractGitHubLanguage()
 
         File.Delete(tempFilePath);
     }
+
+    [WpfFact]
+    public async Task OcrJapaneseImage()
+    {
+        // Given
+        //string testImagePath = jaTestImagePath;
+        string rawOutputFromOCR = await File.ReadAllTextAsync(jaWordBorders);
+
+        HistoryInfo jaHistoryInfo = JsonSerializer.Deserialize<HistoryInfo>(rawOutputFromOCR ?? "[]")
+            ?? throw new Exception("Failed to deserialize HistoryInfo");
+        string expectedResult = jaExpectedResult;
+
+        List<WordBorderInfo> wordBorders = JsonSerializer.Deserialize<List<WordBorderInfo>>(jaHistoryInfo.WordBorderInfoJson ?? "[]")
+            ?? throw new Exception("Failed to deserialize WordBorderInfo list");
+
+        // When
+        GlobalLang japaneseLanguage = new("ja-JP");
+        string ocrTextResult = PostOcrUtilities.GetTextFromWordBorderInfo(wordBorders, japaneseLanguage);
+
+        // Then
+        Assert.Equal(expectedResult, ocrTextResult);
+    }
 }

Tests/StringMethodTests.cs

@@ -35,14 +35,14 @@
         Assert.Equal(expectedWord, singleWordAtSix);
     }
 
-    private static string multiLineInput = @"Hello this is lots 
+    private static string multiLineInput = @"Hello this is lots
 of text which has several lines
-and some spaces at the ends of line 
+and some spaces at the ends of line
 to throw off any easy check";
 
     [Theory]
     [InlineData("Hello", "", " this ...")]
-    [InlineData("lots", "Hello this is ", " ...")]
+    [InlineData("lots", "Hello this is ", "...")]
     [InlineData("of", "...", " text ...")]
     [InlineData("several", "...h has ", " lines...")]
     public void ReturnPreviewsFromWord(string firstWord, string expectedLeftPreview, string expectedRightPreview)
@@ -429,7 +429,7 @@
         Assert.Equal(expected, inputString.LimitCharactersPerLine(charLimit, spotInLine));
     }
 
     private string actualGuids = """
         97a56312-d8e8-4ca5-87fa-18e35266d31e
         bdc5a5f2-d6ff-403d-a632-f9006387e149
         aeef14aa-9aff-4f0d-8ca5-e5df1b399c20

Tests/Tests.csproj

@@ -62,12 +62,18 @@
     </ItemGroup>
 
     <ItemGroup>
+      <Content Update="Images\ja-黒くろごまのちから.png">
+        <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
+      </Content>
       <Content Update="Images\Table-Complex.png">
         <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
       </Content>
     </ItemGroup>
 
     <ItemGroup>
+      <None Update="TextFiles\ja-word-borders.json">
+        <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
+      </None>
       <None Update="TextFiles\Table-Complex-WordBorders.json">
         <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
       </None>

Text-Grab/Utilities/OcrUtilities.cs

@@ -48,12 +48,18 @@
         }
         else
         {
+            // For CJK languages, we need to filter and merge words intelligently
+            List<IOcrWord> words = [.. ocrLine.Words];
+
+            // Filter out likely furigana (very small text fragments that are single characters)
+            List<IOcrWord> filteredWords = FilterFurigana(words);
+
             bool isFirstWord = true;
             bool isPrevWordSpaceJoining = false;
 
             Regex regexSpaceJoiningWord = SpaceJoiningWordRegex();
 
-            foreach (IOcrWord ocrWord in ocrLine.Words)
+            foreach (IOcrWord ocrWord in filteredWords)
             {
                 string wordString = ocrWord.Text;
 
@@ -70,12 +76,79 @@
                 isFirstWord = false;
                 isPrevWordSpaceJoining = isThisWordSpaceJoining;
             }
+
+            // Always add newline after processing each OCR line to preserve line boundaries
+            text.AppendLine();
         }
 
         if (DefaultSettings.CorrectToLatin)
             text.ReplaceGreekOrCyrillicWithLatin();
     }
 
+    private static List<IOcrWord> FilterFurigana(List<IOcrWord> words)
+    {
+        if (words.Count == 0)
+            return words;
+
+        // Calculate the median height of all words
+        List<double> heights = [.. words.Select(w => w.BoundingBox.Height).OrderBy(h => h)];
+        double medianHeight = heights.Count > 0 ? heights[heights.Count / 2] : 0;
+
+        // Furigana is typically 30-50% the height of main text
+        double furiganaThreshold = medianHeight * 0.6;
+
+        List<IOcrWord> filteredWords = [];
+
+        for (int i = 0; i < words.Count; i++)
+        {
+            IOcrWord word = words[i];
+
+            // Check if this word is likely furigana based on:
+            // 1. Much smaller height than median
+            // 2. Single character (especially hiragana/katakana)
+            // 3. Positioned above main text (smaller Y value)
+
+            bool isProbablyFurigana = false;
+
+            if (word.BoundingBox.Height < furiganaThreshold)
+            {
+                // This word is significantly smaller - likely furigana
+                // Check if there's a larger word below it (main kanji)
+                bool hasMainTextBelow = false;
+                for (int j = 0; j < words.Count; j++)
+                {
+                    if (i == j) continue;
+                    IOcrWord otherWord = words[j];
+
+                    // Check if this word is below and overlaps horizontally
+                    bool isBelow = otherWord.BoundingBox.Top > word.BoundingBox.Bottom;
+                    bool overlapsHorizontally = !(otherWord.BoundingBox.Right < word.BoundingBox.Left || 
+                                                 otherWord.BoundingBox.Left > word.BoundingBox.Right);
+                    bool isLarger = otherWord.BoundingBox.Height > furiganaThreshold;
+
+                    if (isBelow && overlapsHorizontally && isLarger)
+                    {
+                        hasMainTextBelow = true;
+                        break;
+                    }
+                }
+
+                if (hasMainTextBelow && word.Text.Length <= 2)
+                {
+                    isProbablyFurigana = true;
+                }
+            }
+
+            if (!isProbablyFurigana)
+            {
+                filteredWords.Add(word);
+            }
+        }
+
+        // If we filtered everything, return original to avoid losing data
+        return filteredWords.Count > 0 ? filteredWords : words;
+    }
+
     public static async Task<string> GetTextFromAbsoluteRectAsync(Rect rect, ILanguage language)
     {
         Rectangle selectedRegion = rect.AsRectangle();
@@ -155,7 +228,7 @@
     {
         if (language is WindowsAiLang winAiLang)
         {
             return new WinAiOcrLinesWords(await WindowsAiUtilities.GetOcrResultAsync(scaledBitmap));
         }
 
         if (language is not GlobalLang globalLang)
@@ -323,8 +396,11 @@
 
         bool isSpaceJoiningOCRLang = language.IsSpaceJoining();
 
-        foreach (IOcrLine ocrLine in ocrResult.Lines)
-            ocrLine.GetTextFromOcrLine(isSpaceJoiningOCRLang, text);
+        if (!isSpaceJoiningOCRLang)
+            text.Append(PostOcrUtilities.GetTextFromOcrResult(ocrResult, language));
+        else
+            foreach (IOcrLine ocrLine in ocrResult.Lines)
+                ocrLine.GetTextFromOcrLine(isSpaceJoiningOCRLang, text);
 
         if (language.IsRightToLeft())
             text.ReverseWordsForRightToLeft();
@@ -373,6 +449,23 @@
         return GetStringFromOcrOutputs(await GetTextFromImageAsync(bmp, language));
     }
 
+    public static async Task<string> OcrAbsoluteFilePathWithLanguageAsync(string absolutePath, Language language)
+    {
+        Uri fileURI = new(absolutePath, UriKind.Absolute);
+        FileInfo fileInfo = new(fileURI.LocalPath);
+        RotateFlipType rotateFlipType = ImageMethods.GetRotateFlipType(absolutePath);
+        BitmapImage droppedImage = new();
+        droppedImage.BeginInit();
+        droppedImage.UriSource = fileURI;
+        ImageMethods.RotateImage(droppedImage, rotateFlipType);
+        droppedImage.CacheOption = BitmapCacheOption.None;
+        droppedImage.EndInit();
+        droppedImage.Freeze();
+        Bitmap bmp = ImageMethods.BitmapImageToBitmap(droppedImage);
+        GlobalLang globalLang = new(language);
+        return GetStringFromOcrOutputs(await GetTextFromImageAsync(bmp, globalLang));
+    }
+
     public static async Task<string> GetClickedWordAsync(Window passedWindow, Point clickedPoint, ILanguage OcrLang)
     {
         using Bitmap bmp = ImageMethods.GetWindowsBoundsBitmap(passedWindow);