Reliable reconstruction of missing observations is essential for environmental time-series analysis, particularly for highly volatile air-quality indicators such as PM2.5. Although linear interpolation is widely used for short-gap imputation due to its simplicity and computational efficiency, it does not explicitly regulate slope dynamics and may produce physically implausible transitions in rapidly fluctuating data. This study proposes a percentile-based slope-constrained linear interpolation method that estimates a slope threshold from the empirical distribution of historical first-order differences and applies a sequential constraint during interpolation to prevent unrealistic gradient changes. The approach requires only a single data-driven parameter and maintains linear computational complexity. • Data-driven slope threshold estimated from the percentile distribution of historical first-order differences. • Sequential slope constraint applied to prevent unrealistic gradient transitions during interpolation. • Linear-time method that preserves the simplicity of standard interpolation while improving reconstruction accuracy.

This research compares retrieval performance between two frequency based feature against texture image retrieval. The aim is that to study the retrieval behavior by using two well-known frequency based features, which has a tiny differences of decomposition basis between DCT and DFT, this work come up with the assumption that different decomposing method might give different retrieval result. In this experiment, feature extraction performs straightforwardly by transforming grayscale global textural of each image into frequency domain without any pre-processing, then similarity measurement performs by Euclidean distance method. The result shows that DFT outperforms DCT for overall precision and recall.

A good content-based image retrieval must provide a robustness to the spatial changing problem of an image which causes from different acquisition, such as, zooming image, same image but different color, etc. These mentioned problems is the main cause to degrade the retrieval performance. To achieve the best retrieval result, an optimal feature must be invented to represent an image semantic. At present, instead of using single feature, most image retrieval system utilizes combining features between two low-level features in an image to retain high precision when faces to the mentioned problem. However, the retrieval error still remains. It causes from the combination of two unsuitable features. Thus, choosing two compatible features is another interested research issue in this area. This paper proposed a new feature combination between color correlograms and edge direction histogram (EDH) in order to give precedence to spatial information in an image. Using color correlogram will treat information about spatial color correlation, while EDH provides the geometry information in the case of the same image but different color. Performance evaluation performs by simple calculation like Euclidean Distance between the query image and image in the database. The experimental result shows that using this combining feature yields a good result while compares with the other combining scheme.