Ȩ > Ç¥ÁØÈ °³¿ä > TTAÀÇ Ç¥ÁØÇöȲ
Ç¥ÁعøÈ£ | TTAK.KO-11.0261 | ±¸Ç¥ÁعøÈ£ | |
---|---|---|---|
Á¦°³Á¤ÀÏ | 2019-12-11 | ÃÑÆäÀÌÁö | 24 |
ÇѱÛÇ¥Áظí | ¿µ»ó ÀÌÇظ¦ À§ÇÑ ¿µ»ó ±¸¹® °ËÃâ ¼ÒÇÁÆ®¿þ¾î Ç°ÁúÆò°¡ Áöħ | ||
¿µ¹®Ç¥Áظí | Guidelines for Evaluation of Visual Phrase Detection Software for Image Understanding | ||
Çѱ۳»¿ë¿ä¾à | ¿µ»ó ±¸¹®(Visual Phrase)Àº, ¿µ»ó¼Ó¿¡¼ ÀÏ°üµÈ »óÈ£ °ü°è(À¯Çü, °ø°£, Å©±â ¹× ½ÃÁ¡ °ü°è)¸¦ °®´Â µÎ °³ ÀÌ»óÀÇ ¿µ»ó °´Ã¼ÀÇ ±×·ìÀ» ÀǹÌÇÑ´Ù. ÀÌ·¯ÇÑ ¿µ»ó ±¸¹®Àº ¿µ»óÀ» ÀÌÇØÇϱâ À§ÇÑ Áß¿äÇÑ ´Ü°èÀ̸ç, ÀÌ°ÍÀ» ÀÌ¿ëÇÏ¸é ¿µ»óÀ» º¸´Ù ÀÇ¹Ì ÀÖ°Ô ±â¼ú(description)Çϰųª °ü½ÉÀÌ ÀÖ´Â ¿µ»ó °´Ã¼¸¦ º¸´Ù »¡¸® ÃßÃâÇÏ°í, ¿µ»óÀ» Çؼ®Çϴµ¥ ÀÖ¾î¼ º¸´Ù Á¤È®ÇÑ °á°ú¸¦ ¾òÀ» ¼ö ÀÖ´Ù.
¿µ»ó ±¸¹® °ËÃâ ±â¼úÀº, ¿À·£ ½Ã°£¿¡ °ÉÃÄ ¸¹Àº ¿¬±¸°¡ ÁøÇàÀÌ µÇ¾î ¿ÔÀ¸¸ç, ±× ¹æ¹ý°ú °á°úµµ ÀÀ¿ëºÐ¾ß¿¡ µû¶ó ¸Å¿ì ´Ù¾çÇÏ´Ù. ±×·¯³ª ¾ÆÁ÷ °³¹ßµÈ ±â¼úÀÌ ½ÇÁ¦ ½Ç¼¼°è¿¡ Àû¿ëÀÌ µÈ °æ¿ì´Â ±ØÈ÷ Èñ¹ÚÇѵ¥, À̴ ó¸®ÇÏ´Â ¿µ»ó µ¥ÀÌÅÍ°¡ ¿ö³« ¹æ´ëÇÏ°í À̸¦ ó¸®ÇÒ ¼ö ÀÖ´Â ±â¼ú(ÄÄÇ»ÅÍ H/W ¹× ¾Ë°í¸®Áò µîÀÇ S/W)ÀÇ ±â¼úÀû ÇÑ°è µîÀ» µé ¼ö ÀÖ´Ù. ±×·¯³ª ÃÖ±Ù¿¡ µîÀåÇÑ µö ·¯´×(Deep Learning) ±â¼úÀº, Àü¿ë H/W, ´ÙÃþ ½Å°æ¸Á ±¸Á¶¿Í ´Ù¾çÇÑ ¾Ë°í¸®ÁòÀÇ µîÀå°ú ±â¼úÀ» Àû¿ëÇÒ ¼ö ÀÖ´Â ´ë¿ë·®ÀÇ µ¥ÀÌÅÍ(big data)¸¦ È¿°úÀûÀ¸·Î È°¿ëÇÏ¿© ¿µ»ó µîÀÇ ºñÁ¤Çüµ¥ÀÌÅ͸¦ º¸´Ù ºü¸£°í È¿À²ÀûÀ¸·Î ó¸®ÇÒ ¼ö ÀÖ°Ô µÇ¾ú´Ù. ¿µ»ó ±¸¹® °ËÃâ ±â¼úÀº, ¿µ»óÀÌ Æ÷ÇÔÇÏ°í ÀÖ´Â ´Ù¾çÇÏ°í dzºÎÇÑ Á¤º¸¸¦ ÃßÃâÇÏ°í È°¿ëÇϱâ À§ÇÑ ÇÙ½É ±â¼ú·Î À̸¦ ÀÌ¿ëÇÑ ÀÚµ¿Â÷ ÁÖÇàÁß Àü¹æ °¨½Ã, ¿µ»óÀÇ ÀÚµ¿ ÁÖ¼® ´Þ±â, µ¿¿µ»óÀÇ °Ë»ö, ¼îÇÎ °ø°£¿¡¼ÀÇ ºÎÁ¤ÇàÀ§ °ËÃâ(fraud detection) µî ´Ù¾çÇÑ ºÐ¾ß¿¡ Àû¿ëÇÒ ¼ö ÀÖ´Ù. º» Ç¥ÁØ¿¡¼´Â µ¿Â÷ ÁÖÇàÁß »ç°í ¹æÁö¸¦ À§ÇÑ Àü¹æ °¨½Ã¸¦ ¸ñÀûÀ¸·Î, Àü¹æ Ä«¸Þ¶ó¸¦ ÅëÇؼ ÀÔ·ÂµÈ ¿µ»óÀ» Çؼ®ÇÏ¿© ¿µ»ó¼Ó °´Ã¼(»ç¶÷, ÀÚÀü°Å, ¹ÙÀÌÅ©, ű º¸µå µî)¸¦ ÀνÄÇÏ°í, À̵éÀÇ °ü°è¸¦ °ËÃâÇÏ´Â ¿µ»ó ±¸¹®(Visual Phrase) °ËÃâ±â¼ú°ú ÀÌ ±â¼úÀÇ ¼º´ÉÀ» °³°üÀûÀÌ°í ü°èÀûÀ¸·Î Æò°¡Çϱâ À§ÇÑ Ç¥ÁØÈµÈ ½ÇÇè ¿µ»ó°ú Æò°¡ ±âÁØ ¹× ü°è¸¦ Á¦°øÇÏ´Â µ¥ ¸ñÀûÀÌ ÀÖ´Ù. |
||
¿µ¹®³»¿ë¿ä¾à | Visual phrase refers to a group of two or more image objects that have a consistent correlation (type, space, size, and viewpoint relationship) in the image. This visual phrase of image is an important step for understanding the image, and by using it, it is possible to describe the image more meaningfully, extract the object of interest more quickly, and obtain more accurate results in interpreting the image.
By introducing visual phrase detection software, which is a core technology for extracting and utilizing various rich information contained in images, we can develop many kind of applications, such as front monitoring during driving, automatic annotation of images, searching for videos, detection of fraud in a shopping space detection and so on. In the standard, the criteria for selection of standard image classes used for the evaluation of visual phrase detection software are defined as the guideline for objectively verifying and evaluating the performance of related technologies. |
||
±¹Á¦Ç¥ÁØ | |||
°ü·ÃÆÄÀÏ | TTAK.KO-11.0261.pdf |