پديد آورندگان :
ﺧﺰاﻋﯽ ﻓﺪاﻓﻦ، ﻣﺴﻌﻮد داﻧﺸﮕﺎه ﻓﻨﯽ و ﺣﺮﻓﻪاي - ﮔﺮوه ﻣﻬﻨﺪﺳﯽ ﺑﺮق، ﺗﻬﺮان، اﯾﺮان , ﻣﻬﺮﺷﺎد، ﻧﺎﺻﺮ داﻧﺸﮕﺎه ﺑﯿﺮﺟﻨﺪ - داﻧﺸﮑﺪه ﻣﻬﻨﺪﺳﯽ ﺑﺮق و ﮐﺎﻣﭙﯿﻮﺗﺮ، ﺑﯿﺮﺟﻨﺪ، اﯾﺮان , رﺿﻮي، ﻣﺤﻤﺪ داﻧﺸﮕﺎه ﺑﯿﺮﺟﻨﺪ - داﻧﺸﮑﺪه ﻣﻬﻨﺪﺳﯽ ﺑﺮق و ﮐﺎﻣﭙﯿﻮﺗﺮ، ﺑﯿﺮﺟﻨﺪ، اﯾﺮان
كليدواژه :
اﺳﺘﺨﺮاج وﯾﮋﮔﯽ , ﺳﯿﺴﺘﻢ ﺑﯿﻨﺎﯾﯽ اﻧﺴﺎن , ﻣﺪل ﻣﺤﺎﺳﺒﺎﺗﯽ ﺳﻠﻮل ﺳﺎده , ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ
چكيده فارسي :
در اﯾﻦ ﻣﻄﺎﻟﻌﻪ، اﻟﮕﻮرﯾﺘﻤﯽ ﻣﻮﺛﺮ و ﮐﺎرآﻣﺪ ﺑﺮاي ﺗﺸﺨﯿﺺ ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ ﺗﺼﻮﯾﺮ ﺑﺮ اﺳﺎس ﻣﺪلﺳﺎزي ﭘﺎﺳﺦ ﺳﺮﯾﻊ ﺳﯿﺴﺘﻢ ﺑﯿﻨﺎﯾﯽ اﻧﺴﺎن ﺑﻪ ﺗﻐﯿﯿﺮات ﺷﺪت روﺷﻨﺎﺋﯽ، ﺑﺎﻓﺖ و رﻧﮓ اراﺋﻪ ﺷﺪه اﺳﺖ. ﺑﺮﺧﯽ ﻣﻮارد ﻣﺎﻧﻨﺪ اﻟﻬﺎم ﮔﺮﻓﺘﻦ از ﻋﻤﻠﮑﺮد ﺳﯿﺴﺘﻢ ﺑﯿﻨﺎﯾﯽ اﻧﺴﺎن، ﻋﺪم ﻧﯿﺎز ﺑﻪ آﻣﻮزش، ﮐﺎﻫﺶ ﺗﻌﺪاد رﻧﮓ، ﮐﺎﻫﺶ ﮐﺎﻧﺎلﻫﺎي رﻧﮕﯽ و اﺳﺘﻔﺎده ﺻﺤﯿﺢ از ﺣﺪاﻗﻞ اﻃﻼﻋﺎت ﺑﺎﻓﺖ در ااﻟﮕﻮرﯾﺘﻢ ﺑﺎﻋﺚ اﻓﺰاﯾﺶ ﮐﺎراﯾﯽ آن ﺷﺪه اﺳﺖ. در روش ﭘﯿﺸﻨﻬﺎدي در ﻣﺮﺣﻠﻪ اول، ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﺣﺴﺎﺳﯿﺖ ﺳﯿﺴﺘﻢ ﺑﯿﻨﺎﯾﯽ اﻧﺴﺎن ﺑﻪ ﺳﯿﮕﻨﺎلﻫﺎي ﺑﺎ ﮐﻨﺘﺮاﺳﺖ ﺑﺎﻻﺗﺮ، ﻓﻘﻂ ﮐﺎﻧﺎل ﺑﺎ ﮐﻨﺘﺮاﺳﺖ ﺑﺎﻻﺗﺮ ﺑﺮاي اﺳﺘﺨﺮاج ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ رﻧﮓ اﺳﺘﻔﺎده و ﺳﭙﺲ ﺑﺎ اﺳﺘﻔﺎده از ﻣﻮﻟﻔﻪ ﺷﺪت روﺷﻨﺎﯾﯽ در ﻓﻀﺎي رﻧﮓ Lab و ﺑﺎ اﺳﺘﻔﺎده از ﻣﺪل ﻣﺤﺎﺳﺒﺎﺗﯽ ﺳﻠﻮل ﺳﺎده ﮐﻮرﺗﮑﺲ ﺑﯿﻨﺎﯾﯽ ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ ﺷﺪت روﺷﻨﺎﺋﯽ و ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ ﺑﺎﻓﺖ اﺳﺘﺨﺮاج ﻣﯽﺷﻮﻧﺪ. در ﻧﻬﺎﯾﺖ، ﺑﺎ ﺗﺮﮐﯿﺐ ﻧﻘﺸﻪﻫﺎي ﺑﺮﺟﺴﺘﮕﯽ رﻧﮓ، ﺷﺪت روﺷﻨﺎﺋﯽ و ﺑﺎﻓﺖ، ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ ﺑﻪدﺳﺖ ﻣﯽآﯾﺪ. روش ﭘﯿﺸﻨﻬﺎدي و روشﻫﺎي ﻣﻮﺟﻮد ﺑﺮروي ﭘﺎﯾﮕﺎه دادهﻫﺎي MSRA10K و ECSSD آزﻣﺎﯾﺶ ﺷﺪه اﺳﺖ. ﻧﺘﺎﯾﺞ ﭘﯿﺎدهﺳﺎزيﻫﺎ ﻧﺸﺎن ﻣﯽدﻫﺪ ﮐﻪ اﻟﮕﻮرﯾﺘﻢ ﺗﺮﮐﯿﺒﯽ ﭘﯿﺸﻨﻬﺎدي ﺑﺮاي ﺗﺸﺨﯿﺺ ﻧﻘﺸﻪ ﺑﺮﺟﺴﺘﮕﯽ ﺑﺎ اﺳﺘﻔﺎده از وﯾﮋﮔﯽﻫﺎي رﻧﮓ و ﺑﺎﻓﺖ ﻏﺎﻟﺐ، در ﭘﺎﯾﮕﺎه داده ECSSD ﺑﻪ ﺗﺮﺗﯿﺐ داراي ﻣﯿﺎﻧﮕﯿﻦ ﺧﻄﺎي ﻣﻄﻠﻖ، اﻣﺘﯿﺎز ﻣﻌﯿﺎر F و ﺳﻄﺢ زﯾﺮ ﻣﻨﺤﻨﯽ ROC ، 0/173 ، 0/789 و 0/891 و در ﭘﺎﯾﮕﺎه داده MSRA10K ﺑﻪ ﺗﺮﺗﯿﺐ 0/790 ،0/178 و 0/919 اﺳﺖ ﮐﻪ در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﺳﺎﯾﺮ ﻣﺪلﻫﺎ ﺑﯿﺎﻧﮕﺮ ﻋﻤﻠﮑﺮد ﺑﻬﺘﺮ روش ﭘﯿﺸﻨﻬﺎدي ﻧﺴﺒﺖ ﺑﻪ ﺳﺎﯾﺮ روشﻫﺎ اﺳﺖ.
چكيده لاتين :
In this study, an effective and efficient algorithm for detection a saliency map is presented based on the modeling of the rapid response of the human visual system to changes in the intensity, texture and color. Some cases such as inspiration from performance of human visual system, requiring no training, reduce number of image colors, reduce color channels and Proper use of the least texture information in this algorithm have increased its efficiency. In the proposed method in the first step , Due to sensitivity of the human visual system to higher contrast signals, only higher contrast channel has been used to extract the color saliency map, Then the intensity saliency map as well as the texture saliency map are extracted using the intensity component in lab color space using Simple cell computational model of the visual cortex and finally, with the perfect combination of the saliency maps of the color, the intensity, and the texture, object saliency map is obtained. The proposed method and existing methods have been tested on MSRA10K and ECSSD databases. The results of the implementations show that the proposed hybrid algorithm for the detection saliency map using the dominant color and texture features, On the ECSSD database, the mean absolute error, F-measure score and the area under the ROC curve are 0.173, 0.789 and 0.891, respectively, and on the MSRA10K database are 0.178, 0.790 and 0.919, respectively, compared to other models, it indicates better performance of the proposed method than other methods.
