“Inpainting” imej adalah satu cabang sains imej yang mampu membina semula bahagian imej yang hilang dan mengembalikan bentuk asalnya. Cabaran terbesar “inpainting” imej adalah untuk mengisi bahagian imej yang hilang dengan cara yang dapat mengekalkan kualiti visual asal imej tersebut. Masalah ini sering ditangani dengan dua jenis algoritma yang bernama “sintesis tekstur” dan “teknik berdasarkan resapan imej”. Algoritma pertama berkesan terhadap imej bercorak dua dimensi (tekstur) manakala algoritma kedua tertumpu kepada struktur linear yang boleh dianggap sebagai imej bercorak satu dimensi, seperti garis dan kontur objek. Cara sintesis tekstur berasaskan contoh imej mengandungi proses penting yang diperlukan untuk meniru tekstur dan struktur imej. Walau bagaimanapun, kejayaan pembiakan struktur imej masih amat bergantung kepada keutamaan susunan pengisian imej berjalan. Algoritma yang menggabungkan kelebihan kedua-dua pendekatan ini telah digunakan dalam applikasi pemulihan imej dan penyingkiran objek dalam imej. Algoritma tersebut merambat nilai keyakinan dalam pixel yang telah disintesiskan sama seperti cara penyebaran maklumat dalam inpainting. Kecekapan pengiraan kod pengaturcaraan dicapai melalui proses persampelan berdasarkan blok. Satu platform interaktif telah ditambah terhadap algoritma tersebut bagi membolehkan pengguna mewujudkan imej topeng yang diperlukan dalam proses “inpainting” dengan cara dinamik dan juga mampu melaksanakan proses “inpainting” berulang kali. Sepuluh gambar telah dipilih sebagai gambar ujian supaya dapat mengesahkan fungsi algoritma yang dicadangkan. Analisis dan perbandingan keputusan gambar ujian yang diperolehi telah dijalankan. Keputusan ujian menunjukkan bahawa algoritma yang dicadangkan berkesan terhadap aplikasi penyingkiran objek dan dapat melaksanakan operasi pemulihan imej pada kebanyakan gambar ujian.
_______________________________________________________________________________________________________
Image inpainting is a branch of image science which is able to reconstruct the missing portions of images and restore its original form. Biggest challenge of image inpainting was to fill in the missing portions that is left behind in a visually plausible way. Often, this problem has been addressed by two classes of algorithms that is “texture synthesis” algorithm and “diffusion-based” techniques. The former works well for repeating two dimensional patterns (textures) while the latter focuses on linear structures which can be thought of as one dimensional patterns, such as lines and object contours. Note that exemplar-based texture synthesis contains the essential process required to replicate both texture and structure. However, the success of structure propagation is highly dependent on the order in which the filling proceeds. Algorithm that combines the advantages of these two approaches was adopted and implemented in the application of image restoration and object removal. The algorithm propagates confidence values in the synthesized pixel in a manner similar to the propagation of information in inpainting. Computational efficiency is achieved by a block based sampling process. An interactive interface was added to the algorithm to enable user create the mask images required in inpainting process dynamically and able of perform inpainting repeatedly. Ten images were chosen as test images so that the proposed algorithm is able to perform its application and the result obtained are analysed and compared. Result shows that the proposed algorithm have an effective effect on object removal and was able to perform image restoration on most of the test images.
Image restoration and object removal using image inpainting / Yeoh Hong Boon