Method analyze

agent/custom/reco/general.py:603–656

(
        self,
        context: Context,
        argv: CustomRecognition.AnalyzeArg,
    )

    """

    def analyze(
        self,
        context: Context,
        argv: CustomRecognition.AnalyzeArg,
    ) -> CustomRecognition.AnalyzeResult | RectType | None:
        try:
            params = parse_params(argv.custom_recognition_param)

            # 获取参数，默认过滤白色
            target_color = params.get("target_color", [255, 255, 255])
            tolerance = params.get("tolerance", 55)
            recognition_node = params.get("recognition")

            if not target_color or len(target_color) != 3:
                logger.error(f"无效的target_color参数: {target_color}")
                return None

            if not recognition_node:
                logger.error("未提供recognition参数")
                return None

            # 获取图像
            img = argv.image

            # 定义目标颜色和颜色容差
            target_color_array = np.array(target_color)

            # 创建颜色过滤掩码
            lower_bound = np.maximum(target_color_array - tolerance, 0)
            upper_bound = np.minimum(target_color_array + tolerance, 255)

            # 创建掩码：保留在目标颜色范围内的像素
            color_mask = np.all((img >= lower_bound) & (img <= upper_bound), axis=-1)

            # 处理图像：目标颜色变成黑色，其他颜色变成白色
            # 创建一个全白图像
            processed_img = np.full_like(img, 255, dtype=np.uint8)
            # 将匹配目标颜色的像素设置为黑色
            processed_img[color_mask] = 0

            # 在处理后的图像上运行OCR识别
            reco_detail = context.run_recognition(recognition_node, processed_img)

            if is_hit(reco_detail):
                logger.debug(f"ColorOCR识别成功: {ocr_text(reco_detail)}")
                return CustomRecognition.AnalyzeResult(
                    box=reco_detail.box, detail=reco_detail.raw_detail
                )
            else:
                return None

        except Exception as e:
            logger.error(f"ColorOCR识别失败: {e}")
            return None


@AgentServer.custom_recognition("ColorOCRWithFallback")