识别结果手工校对

看了一下别的软件校对功能，大多是图片在一边，而识别结果在一边。

可以实现为逐行校对功能，即文字紧挨的就是图片。

先处理界面，增加一个校对按钮，实现处理逻辑代码：

void __fastcall TMainForm::Button_ReviewOCRClick(TObject *Sender) {

    // 识别模块OcrModule：手工校对

    bool checked = Button_ReviewOCR->Down;

    if (checked) {

        int delta = 10;

        int index = 0;

        for (int i = FObjects.size() - 1; i >= 0; --i) {

            TCbwObject * it = FObjects[i];

            TCbwTable * table = dynamic_cast<TCbwTable*>(it);

            if (table) {

                table->Top *= 2;

                table->Zoom(1, 2, table->BeginPoint);

            }

            if (it->Tag == 0)

                continue;

            TText * text = dynamic_cast<TText*>(it);

            if (!text)

                continue;

            TQaOcrObject * ocrObject = (TQaOcrObject*)((void*)(it->Tag));

            ocrObject->RelativeObject = it;

            RECT r = ocrObject->RawRect;

            TCbwImage * imageObject = dynamic_cast<TCbwImage*>(NewMeta(cctImage));

            imageObject->AddPoint(TCbwFloatPoint(r.left, r.top * 2));

            imageObject->AddPoint(TCbwFloatPoint(r.right, r.top + r.bottom));

            imageObject->Tag = int(ocrObject);

            QaImage ocrImage;

            FImage->Crop(r.left - r.left, r.top - r.top, r.right - r.left, r.bottom - r.top, &ocrImage);

            imageObject->CopyFromImage(ocrImage);

            imageObject->AutoSize = true;

            FinishBuild(true);

            int index = IndexOfItemInVector(FObjects, imageObject);

            if (index != -1) {

                FObjects.erase(FObjects.begin() + index);

                FObjects.insert(FObjects.begin(), imageObject);

            }

            it->DragBy(imageObject->Left - it->Left, imageObject->Top + imageObject->Height - it->Top + 10, true);

            ++i;

        }

    }

    else { // 恢复

        cSelectedObjects->Selected = false;

        for (int i = FObjects.size() - 1; i >= 0; --i) {

            TCbwTable * table = dynamic_cast<TCbwTable*>(FObjects[i]);

            if (table) {

                table->Top /= 2;

                table->Zoom(1, 0.5, table->BeginPoint);

            }

            TCbwImage * imageObject = dynamic_cast<TCbwImage*>(FObjects[i]);

            if (!imageObject)

                continue;

            if (imageObject->Tag == 0)

                continue;

            TQaOcrObject * ocrObject = (TQaOcrObject*)((void*)(imageObject->Tag));

            ocrObject->RelativeObject = NULL;

            imageObject->Selected = true;

            cSelectedObjects->AddObject(imageObject);

            DeleteSelectedObjects();

        }

        Application->ProcessMessages();

    }

}

原始识别结果：

校对效果：

这样，在校对时，如果发现有识别错误，可以直接双击文本进行修改。

其实这个功能需求是用户提出来的，他说如果能这样校对的话，他就会大幅提高效率。一想也是哈，直观方便些。

从实现的过程来看，逻辑很简单，不就是找到原始的图片位置，拷贝出对照图片，然后将之与识别结果进行对比排列。

编的软件还是要别人多用，自己闭门造车，只能是自己感觉良好，真正用户喜欢吗？还有待检验。