Method of removing noise using adaptive self-supervised learning

According to an embodiment of the present invention, a noise removal method comprises the steps of: (a) deriving, by a control unit, a first feature map (A1) by performing image processing on an input noise image (A0) using a doughnut kernel passing through a first hidden layer; (b) performing, by t...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: WON KI JEONG, KANGGEUN LEE
Format: Patent
Sprache:eng ; kor
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:According to an embodiment of the present invention, a noise removal method comprises the steps of: (a) deriving, by a control unit, a first feature map (A1) by performing image processing on an input noise image (A0) using a doughnut kernel passing through a first hidden layer; (b) performing, by the control unit, image processing on the first feature map (A1) along a plurality of paths passing through a plurality of hidden layers; (c) performing, by the control unit, concatenating that a plurality of feature maps processed through the paths are combined; (d) determining, by the control unit, whether the steps (a) to (c) are repeatedly re-performed by a set number of times; and (e) displaying, by the control unit, a final image from which noise is removed as a result of determining the re-performing by the set number of times through an output unit. Therefore, a clearer image can be obtained. 본 발명의 실시예에 따른 노이즈 제거방법은 (ㄱ) 제어부가 입력된 노이즈 영상(A0)을 제 1 히든 레이어(Hidden layer 1)를 거치면서 도넛 커널(doughnut Kernel)을 이용한 영상처리로 제 1 특징맵(feature map: A1)을 도출하는 단계; (ㄴ) 상기 제어부가 다수의 히든 레이어를 거치는 다수의 경로를 따라 상기 제 1 특징맵(A1)에 대한 영상 처리를 수행하는 단계; (ㄷ) 상기 제어부가 상기 경로를 거쳐 처리된 다수의 특징맵을 합치는 컨캐터네이팅(Concatenating)을 수행하는 단계; (ㄹ) 상기 제어부는 설정된 횟수로 상기 (ㄱ)단계부터 상기 (ㄷ)단계까지를 반복적으로 재수행한 것인지를 판단하는 단계; 및 (ㅁ) 상기 제어부가 상기 설정된 횟수로 재수행한 판단 결과에 따라 노이즈가 제거된 최종 영상을 출력부를 통해 디스플레이하는 단계;를 포함한다.