Multimedia technology requires network audio technology as well as image processing technology to render acoustic events. Acoustic events include three-dimensional sound image localization. This usually requires three-dimensional directional binaural impulse responses, which are equivalent to head-related transfer functions (HRTFs). This paper describes the sound image projection system, which has inverse filtering for HRTFs in an anechoic space. Inverse filtering for HRTFs, however, is formidable since HRTFs can often be of the non-minimum phase (all-pass). For resolving this problem, we realized the inverse filter of the non-minimum phase component (all-pass component) by using a matched filter with time-delay. From results of computer simulation, we can understand that the inverse filter of HRTFs can be realized by both minimum-phase inverse filter and matched filtering. Moreover, we can success three-dimensional sound image localization using a headphone.