Java中的象形文字或符号识别(Android) [英] Pictograph or symbol recognition in Java (Android)

问题描述

我正在制作一款Android游戏，其核心机制是施法".通过绘制先前指定的符号.我为此尝试了Tesseract，但它将胡言乱语识别为符号，并且没有任何方法(据我所知)来衡量与任何符号的相似性.我可以使用哪种神经网络/OCR实用程序执行任务?

I am making an Android game the core mechanic of which is "spell-casting" by drawing previously specified symbols. I tried Tesseract for this, but it recognizes gibberish as symbols and does not have any way (that I know of) to gauge similarity to any of the symbols. What neural net/ocr utility could I use for my task?

推荐答案

结果证明我需要的是置信度得分.到目前为止，我是从命令行使用Tesseract的.TSV输出选项可以完成此任务:

Turns out what I needed was the confidence score. I used Tesseract from the command line so far. TSV output option does the job:

tesseract test.png stdout --psm 10 tsv

从输出的所有信息中，应注意底部字符串中的最后两列.我使用的符号之一是螺旋形，因此是:螺旋本身和输出.Tesseract将其识别为螺旋形，编码为S，置信度为83，非常好.

From all the information of the output, the attention should be paid to the two last columns in the bottom string. One of the symbols I used is a spiral, so here is the: spiral itself and the output. Tesseract recognized it as a spiral, which is coded as S and the confidence level is 83, which is pretty nice.

另一方面，当我使用乱七八糟的图片时，输出乱码图片仍然显示它是S，但现在的置信度是19.现在，很容易过滤掉无效的输入.

When I use a gibberish picture, on the other hand, the output for gibberish picture still shows it's an S, but now the confidence score is 19. Now it's pretty easy to filter out invalid input.

这篇关于Java中的象形文字或符号识别(Android)的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！