Abstract

Reduced  output  quality  and  being  unaware  of content  are  among  major  issues  with  traditional  image compression techniques. Such issues cause some critical problems when  it  comes  to  quality-intensive  applications,  including object/face  detection and recognition,  Web-based image viewers and  management  systems,  etc.  On  the  other  side,  efficiency  of Web-based  image search engines and retrieval systems in terms of  user  experience  and  usability  could  be  affected.  In  order  to cope with these challenges, a novel image compression method is proposed  that  takes  advantages  of  collective  human  cognitive intelligence to detect the salient object(s) based on  the recognized key concept(s). Then, other less-important regions/objects will be subject  to  the  safe  compression.  Such  an  approach,  besides preserving  semantic  aspects  of  the  images  that  will  result  in smart  (concept-aware)  compression,  could  provide  some crowdsourced  labels  for  more  efficient  indexing  and  annotating of  images.  In  this  regard,  two  birds  could  be  beaten  with  one stone:  compressing  Web  images  with  respect  to  their content/concept  and  annotating  them  with  crowd-suggested labels.  The  experimental  results  as  well  as  user  acceptance evaluation proved the efficacy of the introduced method.