One　of　the　most　challenging　research　issues　in　content-based　image　retrieval　(CBIR)　is　how　to　bridge　the　significant　semantic　gap　between　the　low-level　image　features　and　the　high-level　semantic　concepts.　The　well-known　solutions　are　relevance　feedback　and　regions　of　interest　(ROIs)　detection;　however　both　are　subjective　and　time-consuming.　We　propose　the　visual　information　is　a　new　feature　that　can　objectively　interpret　the　high-level　concepts　and　effectively　reduce　the　semantic　gap　in　image　retrieval.　We　also　make　a　survey　on　the　research　progresses　and　key　technologies　of　visual　perception.　The　research　issues　of　image　retrieval　based　on　visual　perception　are　introduced　as　well　from　four　aspects:　ROIs　detection,　image　segmentation,　relevance　feedback　and　personalized　retrieval.