基于卷積神經(jīng)網(wǎng)絡(luò)的自然場景中數(shù)字的識(shí)別
發(fā)布時(shí)間：2018-07-18 18:50
【摘要】：伴隨著人類社會(huì)步入大數(shù)據(jù)時(shí)代,越來越多的多媒體數(shù)據(jù)涌入互聯(lián)網(wǎng)中,面對海量的圖片數(shù)據(jù),人們迫切的希望可以利用計(jì)算機(jī)來自動(dòng)識(shí)別處理這些多媒體數(shù)據(jù),這也推動(dòng)了計(jì)算機(jī)視覺這一領(lǐng)域的發(fā)展,其中從復(fù)雜背景的圖片中提取文本信息一直是計(jì)算機(jī)視覺中的一個(gè)熱點(diǎn)、難點(diǎn)。近年來神經(jīng)網(wǎng)絡(luò)在計(jì)算機(jī)視覺的各個(gè)方向都獲得了突破性的進(jìn)展,原因是相比于傳統(tǒng)的人工提取圖像特征的方式,神經(jīng)網(wǎng)絡(luò)最大的優(yōu)勢是可以自動(dòng)提取高層特征,這在處理自然場景等復(fù)雜問題中尤其重要,而卷積神經(jīng)網(wǎng)絡(luò)又因自身結(jié)構(gòu)的特點(diǎn)避免了處理圖像這種高維數(shù)據(jù)帶來的計(jì)算量的指數(shù)增長。因此使用卷積神經(jīng)網(wǎng)絡(luò)進(jìn)行場景文本識(shí)別的研究也越來越成為主流。在這樣的背景下,本文的整體思路是將自然場景下的數(shù)字識(shí)別分為字符定位和字符識(shí)別兩個(gè)任務(wù),首先利用卷積神經(jīng)網(wǎng)絡(luò)實(shí)現(xiàn)字符區(qū)域的定位,在獲得字符區(qū)域的準(zhǔn)確位置后,利用循環(huán)神經(jīng)網(wǎng)絡(luò)對該區(qū)域包含的字符串進(jìn)行識(shí)別。在字符定位任務(wù)中,本文通過對物體檢測任務(wù)和自然場景下字符識(shí)別任務(wù)的分析與對比,將目前在物體檢測領(lǐng)域的一個(gè)主流框架Faster-RCNN應(yīng)用于字符定位任務(wù)中,將字符串當(dāng)作一個(gè)特殊的物體。在應(yīng)用Faster-RCNN框架時(shí)針對字符識(shí)別任務(wù)對框架的輸出、網(wǎng)絡(luò)規(guī)模、Anchor比例和IOU閾值等幾個(gè)方面做了優(yōu)化。在字符識(shí)別任務(wù)中,本文使用卷積網(wǎng)絡(luò)和循環(huán)網(wǎng)絡(luò)融合的網(wǎng)絡(luò)結(jié)構(gòu),用卷積網(wǎng)絡(luò)提取特征,用循環(huán)網(wǎng)絡(luò)生成最終的字符序列。分別訓(xùn)練這兩個(gè)部分的網(wǎng)絡(luò),組成一個(gè)完整的識(shí)別系統(tǒng),并在幾個(gè)公開數(shù)據(jù)集上進(jìn)行驗(yàn)證,最后在字符定位的精度方面獲得了優(yōu)于其他方法的效果。
[Abstract]:With the human society stepping into the era of big data, more and more multimedia data pour into the Internet. In the face of massive picture data, people are eager to use computers to automatically identify and process these multimedia data. This also promotes the development of computer vision, in which extracting text information from images with complex backgrounds has been a hot and difficult point in computer vision. In recent years, neural networks have made a breakthrough in all directions of computer vision. The reason is that compared with traditional methods of extracting image features manually, neural networks have the greatest advantage of automatically extracting high-level features. This is particularly important in dealing with complex problems such as natural scenes, and convolution neural networks avoid exponential growth in computation resulting from processing high-dimensional data such as images because of their own structural characteristics. Therefore, the research of scene text recognition based on convolution neural network is becoming more and more popular. In this background, the whole idea of this paper is to divide the digital recognition in the natural scene into two tasks: character location and character recognition. Firstly, we use convolution neural network to locate the character region, and get the exact location of the character region. Cyclic neural network is used to identify the string contained in the region. In the task of character location, by analyzing and comparing the object detection task and the character recognition task in natural scene, this paper applies Faster-RCNN, a mainstream framework in the field of object detection, to the character location task. Treat a string as a special object. In the application of Faster-RCNN framework, several aspects such as the output of character recognition task, the scale of network Anchor ratio and the threshold of IOU are optimized. In the task of character recognition, we use convolutional network and cyclic network structure to extract the feature and generate the final character sequence by using the convolutional network. The network of these two parts is trained to form a complete recognition system, and verified on several open data sets. Finally, the accuracy of character localization is better than that of other methods.
【學(xué)位授予單位】：南京郵電大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41;TP183
本文編號：2132659
[Abstract]:With the human society stepping into the era of big data, more and more multimedia data pour into the Internet. In the face of massive picture data, people are eager to use computers to automatically identify and process these multimedia data. This also promotes the development of computer vision, in which extracting text information from images with complex backgrounds has been a hot and difficult point in computer vision. In recent years, neural networks have made a breakthrough in all directions of computer vision. The reason is that compared with traditional methods of extracting image features manually, neural networks have the greatest advantage of automatically extracting high-level features. This is particularly important in dealing with complex problems such as natural scenes, and convolution neural networks avoid exponential growth in computation resulting from processing high-dimensional data such as images because of their own structural characteristics. Therefore, the research of scene text recognition based on convolution neural network is becoming more and more popular. In this background, the whole idea of this paper is to divide the digital recognition in the natural scene into two tasks: character location and character recognition. Firstly, we use convolution neural network to locate the character region, and get the exact location of the character region. Cyclic neural network is used to identify the string contained in the region. In the task of character location, by analyzing and comparing the object detection task and the character recognition task in natural scene, this paper applies Faster-RCNN, a mainstream framework in the field of object detection, to the character location task. Treat a string as a special object. In the application of Faster-RCNN framework, several aspects such as the output of character recognition task, the scale of network Anchor ratio and the threshold of IOU are optimized. In the task of character recognition, we use convolutional network and cyclic network structure to extract the feature and generate the final character sequence by using the convolutional network. The network of these two parts is trained to form a complete recognition system, and verified on several open data sets. Finally, the accuracy of character localization is better than that of other methods.
【學(xué)位授予單位】：南京郵電大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41;TP183
【參考文獻(xiàn)】
相關(guān)期刊論文 前5條
1 黃凱奇;任偉強(qiáng);譚鐵牛;;圖像物體分類與檢測算法綜述[J];計(jì)算機(jī)學(xué)報(bào);2014年06期
2 劉建偉;劉媛;羅雄麟;;深度學(xué)習(xí)研究進(jìn)展[J];計(jì)算機(jī)應(yīng)用研究;2014年07期
3 余凱;賈磊;陳雨強(qiáng);徐偉;;深度學(xué)習(xí)的昨天、今天和明天[J];計(jì)算機(jī)研究與發(fā)展;2013年09期
4 孫志軍;薛磊;許陽明;王正;;深度學(xué)習(xí)研究綜述[J];計(jì)算機(jī)應(yīng)用研究;2012年08期
5 趙巍;劉家鋒;唐降龍;吳銳;;連續(xù)字符識(shí)別的級聯(lián)HMM訓(xùn)練算法[J];計(jì)算機(jī)學(xué)報(bào);2007年12期
，本文編號：2132659
本文鏈接：http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2132659.html
最近更新
教材專著
