發(fā)布時間：2018-03-27 05:10

  本文選題：光照處理　切入點：圖像拼接　出處：《沈陽工業(yè)大學》2017年碩士論文

【摘要】：圖像拼接技術(shù)被廣泛應(yīng)用于虛擬現(xiàn)實、遙感技術(shù)、醫(yī)學等各個領(lǐng)域,而不同的光照條件對于自然圖像也是必不可少的影響因素,因此不同光照條件下的圖像拼接技術(shù)的研究具有重要學術(shù)意義與應(yīng)用價值。圖像拼接是一種將具有重疊區(qū)域的多張場景圖像拼接融合成一張高分辨率的全景圖像技術(shù),其技術(shù)的關(guān)鍵在于精準地匹配與融合待拼接圖像而使所得結(jié)果圖像沒有拼接跡象。而對于不同光照的圖像預(yù)處理方法則對于圖像拼接效果起到至關(guān)重要的作用,本文對光照不一致情況下的待拼接圖像處理方法和拼接方法進行深入研究以保證較好圖像拼接效果。本文提出了兩種圖像光照處理方法對待拼接圖像進行光照處理。其中逐像素正交分解法提取圖像彩色光照不變信息,去除光照干擾同時保留圖像的彩色信息進行拼接。第二種光照處理方法是基于深度學習框架生成光照恒定圖像,利用深度卷積神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)提取光照信息,通過數(shù)據(jù)訓練模型學習得出平均光照,使待拼接圖像達到光照一致,進而進行下一步的圖像拼接操作。最后對兩種光照處理算法結(jié)果進行對比,并總結(jié)各自的特點與優(yōu)勢。在圖像拼接階段,本文采用重疊區(qū)域匹配驗證方法對圖像誤匹配對進行排除,并通過圓柱投影變換對圖像進行配準拼接,保留圖像原有幾何信息。本文選用波特蘭州立大學的APAP圖像拼接數(shù)據(jù)庫進行圖像拼接實驗,其中深度學習處理光照部分在Ubuntu系統(tǒng)下基于CAFFE網(wǎng)絡(luò)框架完成。實驗證明本文所提方法對于不同光照條件下的圖像拼接取得較好效果。
[Abstract]:Image stitching technology is widely used in virtual reality, remote sensing, medical and other fields, and the effects of different illumination conditions is essential for natural image factors, so according to different research of image stitching technology under the condition of light has important academic significance and application value. Image mosaic is a fusion of the overlapping region a plurality of scene images together into a high resolution panoramic image technology, its key technique is accurate, and fusion images and the results of image stitching and no signs. For different illumination image preprocessing method is a crucial role for image mosaic effect, the light is not consistent to be in-depth studies to ensure better image mosaic method and image mosaic mosaic method under the condition of processing. This paper proposes two kinds of image illumination at Methods the mosaicked image illumination. The pixel color image of orthogonal illumination invariant information decomposition method to remove the light color information of interference while preserving the image mosaic. Second kinds of light treatment method is deep learning framework to generate constant illumination image based on the depth structure of convolutional neural network extraction of light the information obtained by the data, the average light training model of learning, to achieve image mosaic illumination consistency, then the image splicing operation next. At the end of the two light processing algorithm are compared and summed up their own characteristics and advantages. In the stage of image mosaic, overlap area matching verification method to eliminate the image matching error, and through the cylindrical projection transform of image registration, retain original image geometric information. In this paper, the State University of Portland APA P image mosaic database is used for image mosaic experiment. Deep learning processing and illumination part is completed under Ubuntu system based on CAFFE network framework. Experiments show that the proposed method achieves good results for image mosaic under different illumination conditions.

【學位授予單位】：沈陽工業(yè)大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41

【參考文獻】

相關(guān)期刊論文 前3條

1 唐正;劉宏哲;袁家政;;單一光照顏色恒常性計算研究進展[J];計算機科學;2016年11期

2 董俊鵬;楊娟;汪榮貴;曾佳;;基于改進圖像導數(shù)框架的顏色恒常性計算方法[J];計算機工程與應(yīng)用;2016年05期

3 何清;李寧;羅文娟;史忠植;;大數(shù)據(jù)下的機器學習算法綜述[J];模式識別與人工智能;2014年04期

，

本文編號：1670122