Picture Classification Using SVM for Text Segregation

The patent applies segmentation by dividing pictures into two distinct categories: text pictures and non-text pictures. This is achieved through classification based on feature information extraction and Support Vector Machine modeling, which separates the mixed picture collection into homogeneous groups, allowing users to efficiently locate text pictures without sifting through non-text images.

The patent extracts the distinguishing feature information from pictures, specifically focusing on characteristics that indicate whether a picture contains text. By extracting and analyzing feature information such as edge directions and Gabor features, the system identifies and separates text pictures from non-text pictures, effectively extracting the relevant subset of pictures that contain textual content.

The patent applies preliminary action by setting expiration times for stored pictures in advance. When pictures are stored, an expiration time is predetermined and associated with them. The system then automatically monitors and deletes pictures that have exceeded their expiration time, preventing storage space waste while ensuring that pictures within their valid period are preserved.

The patent implements discarding and recovering by automatically identifying and deleting expired pictures from storage. The system recovers storage space by removing pictures that have surpassed their expiration time, while maintaining a mechanism to preserve pictures that are still within their valid storage period, thus optimizing storage resource utilization.

The patent applies local quality by focusing feature extraction on specific local characteristics of pictures that are most indicative of text content. Instead of analyzing all picture features uniformly, the system concentrates on local feature information such as edge directions and Gabor features in regions likely to contain text, thereby maintaining high classification accuracy while reducing overall processing complexity.

The patent employs parameter changes by transforming picture data into specific feature parameter spaces suitable for SVM classification. By converting raw picture data into extracted feature information with specific parameters (edge directions, Gabor features), the system optimizes the input for the classification model, improving accuracy while managing computational requirements through parameter transformation rather than raw data processing.