Ç¥ÁØÈ­ Âü¿©¾È³»

TTAÀÇ Ç¥ÁØÇöȲ

Ȩ > Ç¥ÁØÈ­ °³¿ä > TTAÀÇ Ç¥ÁØÇöȲ

Ç¥ÁعøÈ£ TTAK.KO-10.1473 ±¸Ç¥ÁعøÈ£
Á¦°³Á¤ÀÏ 2023-12-06 ÃÑÆäÀÌÁö 14
ÇѱÛÇ¥ÁØ¸í ºä ÇÕ¼º(View-synthesis)À» ÅëÇÑ ÇнÀ ±â¹Ý ¶óÀÌÆ®Çʵå(Light Field) À̹ÌÁö ¾ÐÃà ÇÁ·¹ÀÓ¿öÅ©
¿µ¹®Ç¥Áظí Learning-based Light Field Image Compression Framework through View-Synthesis
Çѱ۳»¿ë¿ä¾à ÀÌ Ç¥ÁØÀº ½Å°æ¸Á ÇнÀÀ» À§ÇÑ µ¥ÀÌÅͼÂÀ¸·Î ¶óÀÌÆ®Çʵå SAI(sub-aperture image)¸¦ ÀÌ¿ëÇϸç, ¾ÐÃà ÇÁ·¹ÀÓ¿öÅ©´Â Å©°Ô »ùÇøµ °úÁ¤, ºÎ/º¹È£È­±â, »ý¼º ¸ðµ¨ ´ÜÀ¸·Î ÀÌ·ç¾îÁ® ÀÖ´Ù. »ùÇøµ °úÁ¤Àº SAI ¹è¿­¿¡¼­ Ȧ¼ö ȤÀº ¦¼ö Çà°ú ¿­À» ±âÁØÀ¸·Î Å° ºä À̹ÌÁö(Key View)¿Í Å° ºä°¡ ¾Æ´Ñ À̹ÌÁö(Non Key View)¸¦ ¼±Á¤ÇÏ´Â °úÁ¤ÀÌ°í, ºÎº¹È£È­±â´Â È¿°úÀûÀÎ »ó°ü°ü°è Á¦°Å¸¦ À§ÇØ VVC(Versatile Video Coding)¸¦ È°¿ëÇϸç, »ý¼º ¸ðµ¨Àº ¼±Á¤µÈ Å° ºä À̹ÌÁöÀÇ °¢(Angular), °ø°£(Spatial) Á¤º¸¸¦ ÇнÀÇÏ¿© Å° ºä°¡ ¾Æ´Ñ À̹ÌÁö¸¦ º¹¿øÇÑ´Ù.
¿µ¹®³»¿ë¿ä¾à The standard utilizes light Field SAI(sub-aperture image) as an input of neural network training, and the framework is roughly composed with sampling, en/decoder and a generative model. The sampling process is the process of selecting a key view image and a non-key view based on odd or even rows and columns in the SAI array, the en/decoder utilizes Versatile Video Coding (VVC) to effectively eliminate correlations and the generation model learns angular and spatial information of the selected key view image to restore an non key view image.
±¹Á¦Ç¥ÁØ
°ü·ÃÆÄÀÏ TTAK.KO-10.1473.pdf TTAK.KO-10.1473.pdf            

ÀÌÀü
ÀüÀÚÃâÆǹ° ³» Ư¼ö¹®ÀÚ µ¶À½ ¹æ¹ý - Á¦1ºÎ: »ó¼¼ Àбâ
´ÙÀ½
È®ÀåÇö½Ç ÄÜÅÙÃ÷ÀÇ ±³À° ¹× ÈÆ·Ã È¿°ú¼º Æò°¡ ¿ä±¸»çÇ×