To enhance privacy protection for specific regions within an image while preserving its overall visual integrity, this paper presents a visual encryption algorithm targeting partial privacy-sensitive areas. Initially, a novel time-varying delayed exponentially controlled chaotic system (1-TDEC) is proposed. It utilizes the current precision value to modify the system input in a real-time manner, effectively counteracting the dynamic degradation, and making it more suitable for image encryption algorithms designs. Subsequently, the Mask-RCNN instance segmentation model is incorporated to capture the pixel coordinates of different contents in the image, allowing for access to specific area information as required. Furthermore, this encryption algorithm employs a comprehensive permutation process along with parallel bit-shift coupling, effectively transforming the image target areas into visually meaningless forms. It also employs bit-plane theory, providing encryptors the flexibility to set different decryption access levels by manipulating the key stream length from the chaotic system, thus enabling user-specific decryption control. Finally, the comprehensive performance analysis results highlight that the designed algorithm achieves a maximum information entropy value of 7.9998 in encrypted images. Moreover, its resistance to differential attacks (as measured by NPCR and UACI) and pixel correlation are infinitely close to the ideal state, significantly underscoring its security and effectiveness.