freepeople性欧美熟妇, 色戒完整版无删减158分钟hd, 无码精品国产vα在线观看DVD, 丰满少妇伦精品无码专区在线观看,艾栗栗与纹身男宾馆3p50分钟,国产AV片在线观看,黑人与美女高潮,18岁女RAPPERDISSSUBS,国产手机在机看影片

正文內(nèi)容

汽車車牌識別系統(tǒng)-字符識別子系統(tǒng)的設(shè)計(jì)與實(shí)現(xiàn)-免費(fèi)閱讀

2025-09-04 12:22 上一頁面

下一頁面
  

【正文】 as a result, the scale factors and the transform order offer additional keys. On the other hand, note that the phasemask serves as the key of the system, enlarging the key space can be achieved by encoding the target image into two or more phase masks with a modified POCS algorithm. Chang et al have proposed a multiplephases retrieval algorithm and demonstrated that an optical security system based on it has higher level of security and higher quality for the decrypted image. However, this algorithm retrieves only one phasedistribution with a phase constraint in each iteration. As a result, the masks are not so consistent and may affect the quality of the recovered image. In the present paper, we propose a modified POCS algorithm that adjusts the distributions of both phasemasks synchronously in each iteration. As a result, the convergent speed of the iteration process is expected to significantly increase. And the target image with much higher quality is expected to recover because of the coadjusting of the two masks during the iteration process. When the iteration process is finished, the target image is encoded into the phasemasks successfully. Each of these masks severs as the key of the security system and part of the encrypted image itself as well. Moreover, the algorithm can be extended to generate multiple phasemasks for arbitrary stages correlator. To acquire the maximum security, each key is assigned to different authority so that the decryption cannot be performed but being authorized by all of them. This keyassignment scheme is especially useful for military and government applications. The algorithm description is presented in Section 2. Computer simulation of this algorithm and the corresponding discuss are presented in Section 3. 2. Cascaded Iterative Fourier Transform (CIFT) Algorithm Consider the operation of the encryption system with the help of a 4f correlator as shown in , the phase masks placed in the input and the Fourier planes are denoted asand respectively, where (x, y) and (u, v) represent the space and the frequency coordinate, respectively. Once the system is illuminated with a monochromatic plane wave, a target image f(x,y)(an image to be decrypted or verified) is expected to obtain at the output plane. The phasemasksand contain the information of f(x,y), that is,f(x,y)is encoded into these phasemasks. The encoding process is the optimization of the two phasedistributions. It is somewhat similar with the problems of the image reconstruction and the phase retrieval, which can be solved with the POCS algorithm. However, the present problem es down to the phase retrieval in three (or more, in general) planes along the propagation direction. So the conventional POCS algorithm should be modified for this application. The cascaded iteration Fourier transform (CIFT) algorithm begins with the initialization of the phasedistributions of the masks. Suppose the iteration process reaches the kth iteration (k = 1, 2, 3, …), and the phasedistributions in the input and the Fourier plane are represented as and , respectively. Then an estimation of the target image is obtained at the output of the correlator defined by where FT and IFT denote the Fourier transform and the inverse Fourier transform, fk(x,y)satisfies the convergent criterion, the iteration process stops, and and are the optimized distributions. Otherwise, the fk(x,y) is modified to satisfy the target image constraint as follows Then the modified function is transformed backward to generate both of the phasedistributions as follows where ang{ 目前階段,雖然有眾多的研究人員在車牌識別技術(shù)上投入大量的研究,但是由于車牌識別過程存在的偶然性,使得車牌識別的自適應(yīng)性,識別速度,識別率都存在很大的發(fā)展空間。在學(xué)習(xí)階段先要初始化網(wǎng)絡(luò),然后開始輸入要學(xué)習(xí)的樣本,按照網(wǎng)絡(luò)初始設(shè)定的權(quán)重、閾值以及傳輸函數(shù)進(jìn)行計(jì)算,得出每一層神經(jīng)元的輸出,這是從底層向上進(jìn)行的。3.垂直方向數(shù)據(jù)統(tǒng)計(jì)特征提取法這種特征提取方法的算法就是自左向右對圖像進(jìn)行逐列的掃描,統(tǒng)計(jì)每列的黑色的象素的個(gè)數(shù),然后自上而下逐行掃描,統(tǒng)計(jì)每行的黑色象素的個(gè)數(shù),將統(tǒng)計(jì)結(jié)果作為字符的特征向量,如果字符的寬度為w,長度為h,則特征向量的維數(shù)是w+h以上就是幾種基本的特征向量提取方法,還有梯度統(tǒng)計(jì),弧度統(tǒng)計(jì)等其他的特征向量提取方法,另外,還有一種效率極高的特征提取方法-角點(diǎn)提取方法。特征向量的提取方法多種多樣,有逐象素特征提取發(fā),骨架特征提取法,垂直方向數(shù)據(jù)統(tǒng)計(jì)特征提取法,弧度梯度特征提取法等很多種方法,根據(jù)具體情況的不同我們可以來選擇不同的方法。如果環(huán)境是非平穩(wěn)時(shí),神經(jīng)網(wǎng)絡(luò)很難自適應(yīng)學(xué)習(xí)環(huán)境特性;(2) 同時(shí),從數(shù)學(xué)上看BP神經(jīng)網(wǎng)絡(luò)是一個(gè)非線性優(yōu)化算法,存在局部極小問題;(3) 學(xué)習(xí)算法的收斂速度很慢;(4) 對新加入的樣本要影響到已經(jīng)學(xué)完的樣本,刻畫每個(gè)輸入樣本的特征的樹木也要求相同在車牌識別系統(tǒng)中,BP神經(jīng)網(wǎng)絡(luò)利用前面工作中通過車牌定位,二值化,濾波,消除無關(guān)區(qū)域之后,利用投影法得到的分割后的單個(gè)字符經(jīng)過歸一化處理后的字符作為樣本。誤差通過輸出層,按誤差梯度下降的方式修正各層權(quán)值,向隱層、輸入層逐層反傳。在本章首先介紹BP神經(jīng)網(wǎng)絡(luò)的有關(guān)知識,然后利用BP神經(jīng)網(wǎng)絡(luò)解決字符識別的問題。本次畢業(yè)設(shè)計(jì)過程中使用的算法如下:先得到原來字符的高度,跟系統(tǒng)要求的高度做比較,得出要變換的系數(shù),然后根據(jù)得到的系數(shù)求得變換后應(yīng)有得寬度。一種是基于質(zhì)心的位置規(guī)范化;另一種是基于字符外邊框的位置規(guī)范化。如圖37所示: 圖3-5 分割后的圖像我們可以看出分割出來的這幾個(gè)圖像的大小并沒有規(guī)則,有大有小,而在識別的過程中,大小不一的圖片就給識別的過程帶來了一定的麻煩,使得識別的效率和速度大幅度下降,這就需要對圖片進(jìn)行歸一化處理。我們?nèi)∫粋€(gè)略大于7的數(shù)10為閩值,然后統(tǒng)計(jì)每行的非零點(diǎn)數(shù)目。值得慶幸的是由于我們采用的是色彩過濾的方法,所以定位的車牌圖像一般不會(huì)包含太多和牌照區(qū)域顏色不同的邊框。全局閥值方法的優(yōu)點(diǎn)在于算法簡單,對于目標(biāo)和背景明顯分離、直方圖分布呈雙峰的圖像效果良好,但對輸入圖像量化噪聲或不均勻光照等情況抵抗能力差,應(yīng)用受到極大限制。 字符分割的實(shí)現(xiàn)在本次畢業(yè)設(shè)計(jì)的過程中,字符分割所使用的算法為投影法;利用已經(jīng)處理定位的灰色車牌,經(jīng)過一系列的處理,得到車牌圖像的直方圖,利用直方圖的波谷對圖像進(jìn)行分割,然后對分割后的圖像進(jìn)行歸一化處理,最終得到規(guī)格化的分割后的單個(gè)字符圖像,整個(gè)過程如圖33所示 圖3-3 字符分割和歸一化流程167。從圖31和圖32可以看出,字符與字符的分界處往往是投影比較少的地方,并且字符與字符的分界處的投影往往接近于零或者為零,所以去初始閾值t=1對投影圖進(jìn)行掃描,過程如下:1. while(project[i]t) t++,記下位置a;2. while(project[i]=t) i++,記下位置a;3. 得到一個(gè)分割區(qū)域,區(qū)數(shù)加1,重復(fù)步驟(1);4. 如果取數(shù)小于7,則t = t + Δ(自定);5. 重復(fù)1 。去除諸如此類的噪聲是準(zhǔn)確分割字符并保證高識別率的前提。具有數(shù)值計(jì)算和符號計(jì)算、計(jì)算結(jié)果和編程可視化、數(shù)學(xué)和文字統(tǒng)一處理、離線和在線計(jì)算等功能;2. 界面友善、語言自然。MATLAB語言是一種交互性的數(shù)學(xué)腳本語言,其語法與C/C++類似。 本文主要內(nèi)容1. 學(xué)習(xí)MATLAB的使用,熟悉MATLAB函數(shù)的使用;2. 查閱文獻(xiàn),在研究近年來一些典型車牌識別的算法的基礎(chǔ)上,確定進(jìn)行車牌識別的一系列算法。 發(fā)展雖然車牌識別技術(shù)作為智能交通系統(tǒng)中的關(guān)鍵技術(shù),在各國學(xué)者的共同努力下,已經(jīng)得到了長足的發(fā)展,并且已經(jīng)得到了不同程度的實(shí)際應(yīng)用,但目前還存在著種種不足。目前已有的方法很多,但其效果與實(shí)際的要求相差很遠(yuǎn),難以適應(yīng)現(xiàn)代化交 通系統(tǒng)高速度、快節(jié)奏的要求。同時(shí)代,還誕生了面向被盜車輛的第一個(gè)實(shí)時(shí)自動(dòng)車牌監(jiān)測系統(tǒng)。因而從事車牌識別技術(shù)的研究具有極其重要的現(xiàn)實(shí)意義和巨大的經(jīng)濟(jì)價(jià)值。以及利用BP神經(jīng)網(wǎng)絡(luò)的方法規(guī)格化分割出來的字符和模板,在先驗(yàn)條件下識別車牌字符。在不影響汽車運(yùn)行狀態(tài)的情況下,計(jì)算機(jī)自動(dòng)完成車牌的識別,可降低交通管理工作的復(fù)雜度。 BP神經(jīng)網(wǎng)絡(luò) 16167。 引言 9167。 研究現(xiàn)狀 3167。車牌識別技術(shù)綜合了圖形處理、計(jì)算機(jī)視覺、模式識別的技術(shù)以及人工智能等多學(xué)科的知識,其目的就是在無需為車輛加裝其它特殊裝置的情況下實(shí)現(xiàn)對車輛的自動(dòng)監(jiān)測,從而給交通系統(tǒng)的自動(dòng)管理提供了極大的方便,因此車輛牌照自動(dòng)識別系統(tǒng)的實(shí)現(xiàn)是推進(jìn)交通管理計(jì)算機(jī)化的關(guān)鍵技術(shù)之一。在實(shí)現(xiàn)的過程中,字符分割功能主要使用投影法實(shí)現(xiàn),字符識別功能則通過BP神經(jīng)網(wǎng)絡(luò)實(shí)現(xiàn)。 簡介 7167。 字符分割 13167。對此,人們提出了利用交通信息系統(tǒng)、通訊網(wǎng)絡(luò)、定位系統(tǒng)和智能化分析與選線,以緩和道路堵塞和減少交通事故,提高交通利用者的方便、舒適為目的智能交通系統(tǒng),其中汽車車牌識別系統(tǒng)更是智能交通系統(tǒng)的重要組成部分。而國內(nèi)雖有自己的產(chǎn)品,但成本與識別率和識別速度方面卻也是不盡人意,于是車牌識別系統(tǒng)的開發(fā)在國內(nèi)外都有比較大的空間。而在整個(gè)智能交通系統(tǒng)中,車牌識別(License
點(diǎn)擊復(fù)制文檔內(nèi)容
環(huán)評公示相關(guān)推薦
文庫吧 www.dybbs8.com
備案圖鄂ICP備17016276號-1