When game text does not appear immediately (text speed is not the fastest), or when the game has a dynamic background or live2D elements, the captured images will continuously change.
Each time a screenshot is taken, it is compared to the previous screenshot to calculate similarity. When the similarity exceeds the threshold, the image is considered stable, and the next step is performed.
If it can be confirmed that the game is completely static, this value can be set to 0. Conversely, if it isn't, this value should be appropriately increased.
After the image stabilizes, the current image is compared to the image at the last OCR execution (not the last screenshot). When the similarity is lower than this threshold, it is considered that the game text has changed, and OCR is performed.
After each OCR invocation, the current OCR result is compared to the previous OCR result (using edit distance). Only when the edit distance is greater than the threshold will the text be outputted.
This method executes periodically based on the “Execution Interval” and uses the “Text Similarity Threshold” to avoid translating the same text repeatedly.
### Analyze Image Update + Periodic Execution
Combining the above two methods, OCR is executed at least once every “Execution Interval.” It also employs the “Text Similarity Threshold” to avoid translating identical text. Additionally, OCR is performed within intervals based on “Analyze Image Update,” resetting the interval timer.
By default, the following mouse/keyboard events trigger this method: pressing the left mouse button, pressing Enter, releasing Ctrl, releasing Shift, and releasing Alt. If the game window is bound, the method is triggered only when the game window is the foreground window.
After triggering the method, a short wait period is required for the game to render new text, considering that text may not appear immediately (if the text speed is not the fastest).
If the text speed is the fastest, both of the following parameters can be set to 0. Otherwise, the time needed to wait is determined by the following parameters:
This value is similar to the previously mentioned parameter with the same name. However, this is used solely to determine whether the text rendering is complete, and thus it does not share the configuration with the similarly named parameter above.
Due to the unpredictable rendering times of slower text speeds, a specified fixed delay might not suffice. The action is executed when the similarity between the image and the previous screenshot is higher than the threshold.
