DVELOPMENT OF CHARACTER SPOTTING TECHNIQUE AND IMPLEMENTATION TO INTELLIGENT ROBOTS

角色识别技术的发展及其在智能机器人中的实现

• 批准号: 11680403
• 负责人: UMEDA Michio
• 金额: $ 1.54万
• 依托单位: OSAKA ELECTRO-COMMUNICATION UNIVERSITY
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1999
• 资助国家: 日本
• 起止时间: 1999 至 2001
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-11680403/
• 关键词: Character Spotting; Scene Analysis; Image Processing; Character String Extraction; Character Segmentation; Character Recognition; Neural Network; Morphology; 知的ロボット; ハフ変換

In the development of the character spotting technique, we picked up the city scene images which contain the character string information such as bank signboard and proposed the extraction and recognition method of these charactersAs an extraction method of character string, we proposed the dynamic adaptive threshold method based on the connectivity number of picture elements and established the application method to the local target region to perform the high precision. This method is as follows ; the candidate signboared region is extracted from the scene image by using the technique of Hough transform and high performance extraction of character string area is performed by applying the adaptive threshold method ot the local region as a target area. Furthermore applying the mathematical morphology operation to the extracted local area and settling the threshold value based on the pattern spectrum analysis, it is clarified that the highly precise extraction of character string is performed and stable character spotting to rather complex scene images is possibleIn the character recognition process, we proposed the highly precise recognition method by using the autoassociative neural network which learns the shape feature of every character category and applying the network to the character information contained in the bank signboard. This recognition method will be expanded to the object oriented character recognition systemOn the other hand, in the implementation ot the intelligent robots basic control method and some established techniques are implemented to the robots controlled by the radio information

在角色斑点技术的开发中，我们拾取了包含角色字符串信息（例如银行签名板）的城市场景图像，并提出了这些角色的提取和识别方法是字符弦的提取方法，我们提出了基于动态自适应阈值方法，该方法基于连接性的图像元素的连接性数量，并将应用程序元素的应用方法建立到本地目标区域，以执行高精度。该方法如下；通过使用Hough Transform的技术从场景图像中提取候选招牌区域，并通过将自适应阈值方法应用于局部区域作为目标区域来执行角色字符串区域的高性能提取。此外，将数学形态的操作应用于提取的局部区域并基于模式频谱分析来解决阈值的值，澄清，要执行角色字符串的高度精确提取，并且在角色识别过程中可能会在字符识别过程中进行高度清晰的识别方法，从而可以使用自动识别的角色，从而可以在角色识别方面进行自动识别方法，从而可以使用自动识别方法，以实现自动识别方法在银行招牌中。此识别方法将扩展到面向对象的角色识别系统，另一方面，在实现智能机器人基本控制方法和一些已建立的技术中，将实施到由无线电信息控制的机器人

项目成果

Michio, UMEDA: "Handwritten Chinese Charactoer Recognition by Two-stage Composition of Neural Networks"Journal of the Information Processing Society of Japan. Vol.42 No.11. 2724-2732 (2001)

Michio, UMEDA：“通过神经网络的两阶段组合进行手写汉字识别”日本信息处理学会杂志。

梅田三千雄, 横田宏: "ニューラルネットワークの2段階構成による手書き漢字認識"情報処理学会論文誌. 42・11. 2724-2732 (2001)

Michio Umeda、Hiroshi Yokota：“使用两阶段神经网络的手写汉字识别”日本信息处理学会会刊 42・11 2724-2732 (2001)。

Ken-ichi MATSUO, Katsuhiko UEDA and Michio UMEDA: "Extraction of Characer String from Scene Image by Binarizing Local Target Area"Transactions of the Institute of Electrical Engineers of Japan. Vol.122C No.2. 232-241 (2002)

Ken-ichi MATSUO、Katsuhiko UEDA 和 Michio UMEDA：“通过二值化局部目标区域从场景图像中提取字符串”日本电气工程师学会会刊。

梅田三千雄: "個別文字の併用認識による高精度化"大阪電気通信大学ISCテクニカルレポート. 2000-03. 27-32 (2001)

Michio Umeda：“通过单个字符的组合识别提高准确性”大阪电气通信大学 ISC 技术报告 2000-03 (2001)。

Ken-ichi Matsuo et al: "Extraction of Character String Regions from Scene Images Attended to Signboard Region"Proceedings of the 3rd International Workshop on Advanced Image Technology. 171-176 (2000)

Ken-ichi Matsuo 等人：“从涉及招牌区域的场景图像中提取字符串区域”第三届高级图像技术国际研讨会论文集。